{ "created": "2026-10-26T23:20:40.520851+00:00", "scope": "Four authored live API/Qwen/TTS trials; not microphone-to-ear or a human listening rating.", "rows": [ { "case": "cat_joke", "input": "Erzähl mir einen Witz über Katzen.", "history": [], "events": [ { "at_ms": 386.28, "text": "Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun.", "script": "(amusement, playful) Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun.", "language": "de", "performance": { "sentences": [ { "emotion": "amusement", "secondary": null, "blend": 0.0, "intensity": 0.72, "direction": "amusement, playful", "text": "Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun.", "start_s": 0.0, "end_s": 5.1 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "amusement.soft", "sentence_index": 0, "sentence_text": "Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun.", "timing": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 314.79, "llm_first_sentence_ms": 360.13, "sentence_ready_ms": 372.15, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "amusement.soft" } }, { "at_ms": 1528.23, "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 314.79, "llm_first_sentence_ms": 360.13, "finish_reason": "stop", "usage": { "prompt_tokens": 1625, "completion_tokens": 73, "total_tokens": 1698, "cost": 0.0017003448, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00171752, "upstream_inference_prompt_cost": 0.00160875, "upstream_inference_completions_cost": 0.00010877 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 395.57, "response_complete": true, "answer_first_audio_ms": 1514.91, "tts_first_audio_ms": 1142.7, "first_audio_ms": 1514.92 }, { "at_ms": 1528.44, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 2350.42, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 3221.39, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 3803.6, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 21120 }, { "at_ms": 3893.26, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 15120 }, { "at_ms": 3895.91, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 3896.07, "text": "Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun.\nDie eine sagt: „Na und?", "script": "(amusement, playful) Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun. (soft, firm) Die eine sagt: „Na und?", "language": "de", "performance": { "sentences": [ { "emotion": "amusement", "secondary": null, "blend": 0.0, "intensity": 0.72, "direction": "amusement, playful", "text": "Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun.", "start_s": 0.0, "end_s": 5.31 }, { "emotion": "contentment", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "soft, firm", "text": "Die eine sagt: „Na und?", "start_s": 5.31, "end_s": 7.01 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "delivery.firm.soft", "sentence_index": 1, "sentence_text": "Die eine sagt: „Na und?", "timing": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 314.79, "llm_first_sentence_ms": 360.13, "finish_reason": "stop", "usage": { "prompt_tokens": 1625, "completion_tokens": 73, "total_tokens": 1698, "cost": 0.0017003448, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00171752, "upstream_inference_prompt_cost": 0.00160875, "upstream_inference_completions_cost": 0.00010877 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 395.57, "response_complete": true, "answer_first_audio_ms": 1514.91, "tts_first_audio_ms": 1142.7, "first_audio_ms": 1514.92, "tts_server_stages": [ { "generation_ms": 3282, "alignment_ms": 85, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3497, "performance": { "device_seconds": 2.6604978041648866, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4683975007332547, "audio_seconds_per_second": 2.134938803974287, "batches": [ { "batch_size": 1, "device_seconds": 2.6604978041648866, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1097188740968704, "first_audio_seconds": 0.7743006591796875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.4418344423174858 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 3508.68, "audio_s": 5.31 } ], "sentence_ready_ms": 3881.68, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "delivery.firm.soft" } }, { "at_ms": 5019.31, "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 314.79, "llm_first_sentence_ms": 360.13, "finish_reason": "stop", "usage": { "prompt_tokens": 1625, "completion_tokens": 73, "total_tokens": 1698, "cost": 0.0017003448, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00171752, "upstream_inference_prompt_cost": 0.00160875, "upstream_inference_completions_cost": 0.00010877 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 395.57, "response_complete": true, "answer_first_audio_ms": 1514.91, "tts_first_audio_ms": 1142.7, "first_audio_ms": 1514.92, "tts_server_stages": [ { "generation_ms": 3282, "alignment_ms": 85, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3497, "performance": { "device_seconds": 2.6604978041648866, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4683975007332547, "audio_seconds_per_second": 2.134938803974287, "batches": [ { "batch_size": 1, "device_seconds": 2.6604978041648866, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1097188740968704, "first_audio_seconds": 0.7743006591796875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.4418344423174858 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 3508.68, "audio_s": 5.31 } ] }, { "at_ms": 5019.48, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 5547.24, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38160 }, { "at_ms": 5596.71, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 5596.9, "text": "Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun.\nDie eine sagt: „Na und? Was macht die da unten?“", "script": "(amusement, playful) Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun. (soft, firm) Die eine sagt: „Na und? (soft, firm) Was macht die da unten?“", "language": "de", "performance": { "sentences": [ { "emotion": "amusement", "secondary": null, "blend": 0.0, "intensity": 0.72, "direction": "amusement, playful", "text": "Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun.", "start_s": 0.0, "end_s": 5.31 }, { "emotion": "contentment", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "soft, firm", "text": "Die eine sagt: „Na und?", "start_s": 5.31, "end_s": 7.5 }, { "emotion": "contentment", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "soft, firm", "text": "Was macht die da unten?“", "start_s": 7.5, "end_s": 9.2 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "delivery.firm.soft", "sentence_index": 2, "sentence_text": "Was macht die da unten?“", "timing": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 314.79, "llm_first_sentence_ms": 360.13, "finish_reason": "stop", "usage": { "prompt_tokens": 1625, "completion_tokens": 73, "total_tokens": 1698, "cost": 0.0017003448, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00171752, "upstream_inference_prompt_cost": 0.00160875, "upstream_inference_completions_cost": 0.00010877 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 395.57, "response_complete": true, "answer_first_audio_ms": 1514.91, "tts_first_audio_ms": 1142.7, "first_audio_ms": 1514.92, "tts_server_stages": [ { "generation_ms": 3282, "alignment_ms": 85, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3497, "performance": { "device_seconds": 2.6604978041648866, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4683975007332547, "audio_seconds_per_second": 2.134938803974287, "batches": [ { "batch_size": 1, "device_seconds": 2.6604978041648866, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1097188740968704, "first_audio_seconds": 0.7743006591796875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.4418344423174858 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1518, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1693, "performance": { "device_seconds": 1.1939678473472597, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.4663936903700233, "audio_seconds_per_second": 2.1441113390848594, "batches": [ { "batch_size": 1, "device_seconds": 1.1939678473472597, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.3423648327589035, "first_audio_seconds": 0.7263375244140625, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.14281930029392242 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 3508.68, "audio_s": 5.31 }, { "sentence": 1, "total_ms": 1699.91, "audio_s": 2.19 } ], "sentence_ready_ms": 5582.45, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "delivery.firm.soft" } }, { "at_ms": 6744.1, "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 314.79, "llm_first_sentence_ms": 360.13, "finish_reason": "stop", "usage": { "prompt_tokens": 1625, "completion_tokens": 73, "total_tokens": 1698, "cost": 0.0017003448, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00171752, "upstream_inference_prompt_cost": 0.00160875, "upstream_inference_completions_cost": 0.00010877 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 395.57, "response_complete": true, "answer_first_audio_ms": 1514.91, "tts_first_audio_ms": 1142.7, "first_audio_ms": 1514.92, "tts_server_stages": [ { "generation_ms": 3282, "alignment_ms": 85, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3497, "performance": { "device_seconds": 2.6604978041648866, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4683975007332547, "audio_seconds_per_second": 2.134938803974287, "batches": [ { "batch_size": 1, "device_seconds": 2.6604978041648866, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1097188740968704, "first_audio_seconds": 0.7743006591796875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.4418344423174858 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1518, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1693, "performance": { "device_seconds": 1.1939678473472597, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.4663936903700233, "audio_seconds_per_second": 2.1441113390848594, "batches": [ { "batch_size": 1, "device_seconds": 1.1939678473472597, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.3423648327589035, "first_audio_seconds": 0.7263375244140625, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.14281930029392242 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 3508.68, "audio_s": 5.31 }, { "sentence": 1, "total_ms": 1699.91, "audio_s": 2.19 } ] }, { "at_ms": 6744.27, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 7265.14, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 40080 }, { "at_ms": 7345.25, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 7345.49, "text": "Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun.\nDie eine sagt: „Na und? Was macht die da unten?“\nDie andere antwortet: „Ich weiß es nicht, aber ich wette, sie ist gerade sehr beschäftigt.“", "script": "(amusement, playful) Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun. (soft, firm) Die eine sagt: „Na und? (soft, firm) Was macht die da unten?“ (surprise, natural) Die andere antwortet: „Ich weiß es nicht, aber ich wette, sie ist gerade sehr beschäftigt.“", "language": "de", "performance": { "sentences": [ { "emotion": "amusement", "secondary": null, "blend": 0.0, "intensity": 0.72, "direction": "amusement, playful", "text": "Zwei Katzen sitzen auf einem Dach und schauen auf eine kleine Maus unter dem Zaun.", "start_s": 0.0, "end_s": 5.31 }, { "emotion": "contentment", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "soft, firm", "text": "Die eine sagt: „Na und?", "start_s": 5.31, "end_s": 7.5 }, { "emotion": "contentment", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "soft, firm", "text": "Was macht die da unten?“", "start_s": 7.5, "end_s": 9.77 }, { "emotion": "surprise", "secondary": "contentment", "blend": 0.25, "intensity": 0.72, "direction": "surprise, natural", "text": "Die andere antwortet: „Ich weiß es nicht, aber ich wette, sie ist gerade sehr beschäftigt.“", "start_s": 9.77, "end_s": 14.87 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "surprise.soft", "sentence_index": 3, "sentence_text": "Die andere antwortet: „Ich weiß es nicht, aber ich wette, sie ist gerade sehr beschäftigt.“", "timing": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 314.79, "llm_first_sentence_ms": 360.13, "finish_reason": "stop", "usage": { "prompt_tokens": 1625, "completion_tokens": 73, "total_tokens": 1698, "cost": 0.0017003448, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00171752, "upstream_inference_prompt_cost": 0.00160875, "upstream_inference_completions_cost": 0.00010877 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 395.57, "response_complete": true, "answer_first_audio_ms": 1514.91, "tts_first_audio_ms": 1142.7, "first_audio_ms": 1514.92, "tts_server_stages": [ { "generation_ms": 3282, "alignment_ms": 85, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3497, "performance": { "device_seconds": 2.6604978041648866, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4683975007332547, "audio_seconds_per_second": 2.134938803974287, "batches": [ { "batch_size": 1, "device_seconds": 2.6604978041648866, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1097188740968704, "first_audio_seconds": 0.7743006591796875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.4418344423174858 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1518, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1693, "performance": { "device_seconds": 1.1939678473472597, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.4663936903700233, "audio_seconds_per_second": 2.1441113390848594, "batches": [ { "batch_size": 1, "device_seconds": 1.1939678473472597, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.3423648327589035, "first_audio_seconds": 0.7263375244140625, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.14281930029392242 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1546, "alignment_ms": 77, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1741, "performance": { "device_seconds": 1.1861202468872072, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.47827429309968034, "audio_seconds_per_second": 2.0908504061948054, "batches": [ { "batch_size": 1, "device_seconds": 1.1861202468872072, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.3701241239905357, "first_audio_seconds": 0.7434956665039063, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.17879826575517654 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 3508.68, "audio_s": 5.31 }, { "sentence": 1, "total_ms": 1699.91, "audio_s": 2.19 }, { "sentence": 2, "total_ms": 1747.46, "audio_s": 2.27 } ], "sentence_ready_ms": 7330.94, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "surprise.soft" } }, { "at_ms": 8522.22, "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 314.79, "llm_first_sentence_ms": 360.13, "finish_reason": "stop", "usage": { "prompt_tokens": 1625, "completion_tokens": 73, "total_tokens": 1698, "cost": 0.0017003448, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00171752, "upstream_inference_prompt_cost": 0.00160875, "upstream_inference_completions_cost": 0.00010877 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 395.57, "response_complete": true, "answer_first_audio_ms": 1514.91, "tts_first_audio_ms": 1142.7, "first_audio_ms": 1514.92, "tts_server_stages": [ { "generation_ms": 3282, "alignment_ms": 85, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3497, "performance": { "device_seconds": 2.6604978041648866, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4683975007332547, "audio_seconds_per_second": 2.134938803974287, "batches": [ { "batch_size": 1, "device_seconds": 2.6604978041648866, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1097188740968704, "first_audio_seconds": 0.7743006591796875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.4418344423174858 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1518, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1693, "performance": { "device_seconds": 1.1939678473472597, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.4663936903700233, "audio_seconds_per_second": 2.1441113390848594, "batches": [ { "batch_size": 1, "device_seconds": 1.1939678473472597, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.3423648327589035, "first_audio_seconds": 0.7263375244140625, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.14281930029392242 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1546, "alignment_ms": 77, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1741, "performance": { "device_seconds": 1.1861202468872072, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.47827429309968034, "audio_seconds_per_second": 2.0908504061948054, "batches": [ { "batch_size": 1, "device_seconds": 1.1861202468872072, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.3701241239905357, "first_audio_seconds": 0.7434956665039063, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.17879826575517654 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 3508.68, "audio_s": 5.31 }, { "sentence": 1, "total_ms": 1699.91, "audio_s": 2.19 }, { "sentence": 2, "total_ms": 1747.46, "audio_s": 2.27 } ] }, { "at_ms": 8522.4, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 9301.64, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 10159.78, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 10903.24, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 26880 }, { "at_ms": 10980.71, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 22800 }, { "at_ms": 10981.49, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 10981.66, "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 314.79, "llm_first_sentence_ms": 360.13, "finish_reason": "stop", "usage": { "prompt_tokens": 1625, "completion_tokens": 73, "total_tokens": 1698, "cost": 0.0017003448, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00171752, "upstream_inference_prompt_cost": 0.00160875, "upstream_inference_completions_cost": 0.00010877 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 395.57, "response_complete": true, "answer_first_audio_ms": 1514.91, "tts_first_audio_ms": 1142.7, "first_audio_ms": 1514.92, "tts_server_stages": [ { "generation_ms": 3282, "alignment_ms": 85, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3497, "performance": { "device_seconds": 2.6604978041648866, "raw_audio_seconds": 5.68, "aggregate_rtf": 0.4683975007332547, "audio_seconds_per_second": 2.134938803974287, "batches": [ { "batch_size": 1, "device_seconds": 2.6604978041648866, "audio_seconds": [ 5.68 ], "model_wall_seconds": 3.1097188740968704, "first_audio_seconds": 0.7743006591796875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.4418344423174858 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1518, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1693, "performance": { "device_seconds": 1.1939678473472597, "raw_audio_seconds": 2.56, "aggregate_rtf": 0.4663936903700233, "audio_seconds_per_second": 2.1441113390848594, "batches": [ { "batch_size": 1, "device_seconds": 1.1939678473472597, "audio_seconds": [ 2.56 ], "model_wall_seconds": 1.3423648327589035, "first_audio_seconds": 0.7263375244140625, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.14281930029392242 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1546, "alignment_ms": 77, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1741, "performance": { "device_seconds": 1.1861202468872072, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.47827429309968034, "audio_seconds_per_second": 2.0908504061948054, "batches": [ { "batch_size": 1, "device_seconds": 1.1861202468872072, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.3701241239905357, "first_audio_seconds": 0.7434956665039063, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.17879826575517654 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 3421, "alignment_ms": 70, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3623, "performance": { "device_seconds": 2.7614823813438414, "raw_audio_seconds": 5.92, "aggregate_rtf": 0.4664666184702435, "audio_seconds_per_second": 2.143776125458785, "batches": [ { "batch_size": 1, "device_seconds": 2.7614823813438414, "audio_seconds": [ 5.92 ], "model_wall_seconds": 3.245129093527794, "first_audio_seconds": 0.7783905029296875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.4769063740968704 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 3508.68, "audio_s": 5.31 }, { "sentence": 1, "total_ms": 1699.91, "audio_s": 2.19 }, { "sentence": 2, "total_ms": 1747.46, "audio_s": 2.27 }, { "sentence": 3, "total_ms": 3637.09, "audio_s": 5.87 } ], "total_ms": 10968.2, "audio_s": 15.64, "sentences": 4, "generation_rtf": 0.7013 } ], "first_pcm_ms": 1528.44, "first_main_audio_ms": 1528.44, "audio_s": 15.64 }, { "case": "another_joke", "input": "Noch einen bitte.", "history": [ { "role": "user", "content": "Erzähl mir einen Witz über Katzen." }, { "role": "assistant", "content": "Ein bisschen Spaß passt immer. Was bestellt eine Katze im Café? Einen Miauccino!" } ], "events": [ { "at_ms": 342.68, "text": "Warum sitzen Katzen immer genau auf der Tastatur?", "script": "(interest, playful) Warum sitzen Katzen immer genau auf der Tastatur?", "language": "de", "performance": { "sentences": [ { "emotion": "interest", "secondary": "amusement", "blend": 0.25, "intensity": 0.72, "direction": "interest, playful", "text": "Warum sitzen Katzen immer genau auf der Tastatur?", "start_s": 0.0, "end_s": 2.72 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "interest.soft", "sentence_index": 0, "sentence_text": "Warum sitzen Katzen immer genau auf der Tastatur?", "timing": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 301.57, "llm_first_sentence_ms": 327.33, "sentence_ready_ms": 331.48, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "interest.soft" } }, { "at_ms": 1387.57, "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 301.57, "llm_first_sentence_ms": 327.33, "finish_reason": "stop", "usage": { "prompt_tokens": 1663, "completion_tokens": 25, "total_tokens": 1688, "cost": 0.0016667838, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00168362, "upstream_inference_prompt_cost": 0.00164637, "upstream_inference_completions_cost": 3.725e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 349.69, "response_complete": true, "answer_first_audio_ms": 1376.85, "tts_first_audio_ms": 1045.31, "first_audio_ms": 1376.85 }, { "at_ms": 1387.74, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 2222.07, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 2903.64, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 26880 }, { "at_ms": 2982.49, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 17040 }, { "at_ms": 2985.64, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 2985.75, "text": "Warum sitzen Katzen immer genau auf der Tastatur? Weil sie den besten Zugang zum WLAN haben!", "script": "(interest, playful) Warum sitzen Katzen immer genau auf der Tastatur? (interest, playful) Weil sie den besten Zugang zum WLAN haben!", "language": "de", "performance": { "sentences": [ { "emotion": "interest", "secondary": "amusement", "blend": 0.25, "intensity": 0.72, "direction": "interest, playful", "text": "Warum sitzen Katzen immer genau auf der Tastatur?", "start_s": 0.0, "end_s": 4.03 }, { "emotion": "interest", "secondary": "amusement", "blend": 0.25, "intensity": 0.72, "direction": "interest, playful", "text": "Weil sie den besten Zugang zum WLAN haben!", "start_s": 4.03, "end_s": 6.75 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "interest.soft", "sentence_index": 1, "sentence_text": "Weil sie den besten Zugang zum WLAN haben!", "timing": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 301.57, "llm_first_sentence_ms": 327.33, "finish_reason": "stop", "usage": { "prompt_tokens": 1663, "completion_tokens": 25, "total_tokens": 1688, "cost": 0.0016667838, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00168362, "upstream_inference_prompt_cost": 0.00164637, "upstream_inference_completions_cost": 3.725e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 349.69, "response_complete": true, "answer_first_audio_ms": 1376.85, "tts_first_audio_ms": 1045.31, "first_audio_ms": 1376.85, "tts_server_stages": [ { "generation_ms": 2435, "alignment_ms": 72, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2629, "performance": { "device_seconds": 1.9919096102714537, "raw_audio_seconds": 4.32, "aggregate_rtf": 0.4610901875628365, "audio_seconds_per_second": 2.1687731098457217, "batches": [ { "batch_size": 1, "device_seconds": 1.9919096102714537, "audio_seconds": [ 4.32 ], "model_wall_seconds": 2.3124767914414406, "first_audio_seconds": 0.723662841796875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.31441589444875717 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2641.37, "audio_s": 4.03 } ], "sentence_ready_ms": 2973.98, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "interest.soft" } }, { "at_ms": 4153.97, "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 301.57, "llm_first_sentence_ms": 327.33, "finish_reason": "stop", "usage": { "prompt_tokens": 1663, "completion_tokens": 25, "total_tokens": 1688, "cost": 0.0016667838, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00168362, "upstream_inference_prompt_cost": 0.00164637, "upstream_inference_completions_cost": 3.725e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 349.69, "response_complete": true, "answer_first_audio_ms": 1376.85, "tts_first_audio_ms": 1045.31, "first_audio_ms": 1376.85, "tts_server_stages": [ { "generation_ms": 2435, "alignment_ms": 72, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2629, "performance": { "device_seconds": 1.9919096102714537, "raw_audio_seconds": 4.32, "aggregate_rtf": 0.4610901875628365, "audio_seconds_per_second": 2.1687731098457217, "batches": [ { "batch_size": 1, "device_seconds": 1.9919096102714537, "audio_seconds": [ 4.32 ], "model_wall_seconds": 2.3124767914414406, "first_audio_seconds": 0.723662841796875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.31441589444875717 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2641.37, "audio_s": 4.03 } ] }, { "at_ms": 4154.14, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 4933.46, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 5774.93, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 6208.75, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 11520 }, { "at_ms": 6274.36, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 24000 }, { "at_ms": 6275.02, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 6275.13, "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 301.57, "llm_first_sentence_ms": 327.33, "finish_reason": "stop", "usage": { "prompt_tokens": 1663, "completion_tokens": 25, "total_tokens": 1688, "cost": 0.0016667838, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.00168362, "upstream_inference_prompt_cost": 0.00164637, "upstream_inference_completions_cost": 3.725e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "task_acknowledgement_only": false, "task_minimum_units": 2, "task_missing_units": false, "completion_repairs": 0, "llm_total_ms": 349.69, "response_complete": true, "answer_first_audio_ms": 1376.85, "tts_first_audio_ms": 1045.31, "first_audio_ms": 1376.85, "tts_server_stages": [ { "generation_ms": 2435, "alignment_ms": 72, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2629, "performance": { "device_seconds": 1.9919096102714537, "raw_audio_seconds": 4.32, "aggregate_rtf": 0.4610901875628365, "audio_seconds_per_second": 2.1687731098457217, "batches": [ { "batch_size": 1, "device_seconds": 1.9919096102714537, "audio_seconds": [ 4.32 ], "model_wall_seconds": 2.3124767914414406, "first_audio_seconds": 0.723662841796875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.31441589444875717 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 3095, "alignment_ms": 55, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3276, "performance": { "device_seconds": 2.4765242171287536, "raw_audio_seconds": 5.28, "aggregate_rtf": 0.46903867748650635, "audio_seconds_per_second": 2.1320203386185965, "batches": [ { "batch_size": 1, "device_seconds": 2.4765242171287536, "audio_seconds": [ 5.28 ], "model_wall_seconds": 2.9171821251511574, "first_audio_seconds": 0.7567337646484374, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.43392543494701385 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2641.37, "audio_s": 4.03 }, { "sentence": 1, "total_ms": 3290.51, "audio_s": 5.28 } ], "total_ms": 6264.62, "audio_s": 9.31, "sentences": 2, "generation_rtf": 0.6729 } ], "first_pcm_ms": 1387.74, "first_main_audio_ms": 1387.74, "audio_s": 9.31 }, { "case": "bereavement", "input": "Mein Hund ist heute gestorben.", "history": [], "events": [ { "at_ms": 323.43, "text": "Das tut mir sehr leid.", "script": "(contentment, soft, warm) Das tut mir sehr leid.", "language": "de", "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das tut mir sehr leid.", "start_s": 0.0, "end_s": 1.7 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "reaction", "template_id": "contentment.soft", "sentence_index": 0, "sentence_text": "Das tut mir sehr leid.", "timing": { "quick_reply_status": "selected", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "sentence_ready_ms": 313.86, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "contentment.soft" } }, { "at_ms": 860.75, "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "tts_first_audio_ms": 537.82, "first_audio_ms": 851.74 }, { "at_ms": 860.82, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 3590 }, { "at_ms": 1093.06, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 9600 }, { "at_ms": 1331.7, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 9600 }, { "at_ms": 1564.15, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 9600 }, { "at_ms": 1804.02, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 9600 }, { "at_ms": 2046.45, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 9600 }, { "at_ms": 2285.82, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 7680 }, { "at_ms": 2295.67, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 250 }, { "at_ms": 2295.78, "text": "Das tut mir sehr leid. Oh, das tut mir so, so sehr leid.", "script": "(contentment, soft, warm) Das tut mir sehr leid. (sadness, soft) Oh, das tut mir so, so sehr leid.", "language": "de", "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das tut mir sehr leid.", "start_s": 0.0, "end_s": 2.48 }, { "emotion": "sadness", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "sadness, soft", "text": "Oh, das tut mir so, so sehr leid.", "start_s": 2.48, "end_s": 5.2 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "sadness.soft", "sentence_index": 1, "sentence_text": "Oh, das tut mir so, so sehr leid.", "timing": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "tts_first_audio_ms": 537.82, "first_audio_ms": 851.74, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 1240.67, "llm_first_sentence_ms": 1270.93, "finish_reason": "stop", "usage": { "prompt_tokens": 1363, "completion_tokens": 77, "total_tokens": 1440, "cost": 0.001449459, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0014641, "upstream_inference_prompt_cost": 0.00134937, "upstream_inference_completions_cost": 0.00011473 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 1348.05, "response_complete": true, "tts_server_stages": [ { "generation_ms": 1840, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1963, "performance": { "device_seconds": 1.6535211882591248, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.6667424146206148, "audio_seconds_per_second": 1.4998295864663314, "batches": [ { "batch_size": 1, "device_seconds": 1.6535211882591248, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.6557160764932632, "first_audio_seconds": 0.2259435577392578, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.001526370644569397 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.48, "quick_reply_tts_ms": 1968.76, "sentence_tts": [ { "sentence": 0, "total_ms": 1968.77, "audio_s": 2.48 } ], "sentence_ready_ms": 2284.44, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "sadness.soft" } }, { "at_ms": 3482.78, "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "tts_first_audio_ms": 537.82, "first_audio_ms": 851.74, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 1240.67, "llm_first_sentence_ms": 1270.93, "finish_reason": "stop", "usage": { "prompt_tokens": 1363, "completion_tokens": 77, "total_tokens": 1440, "cost": 0.001449459, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0014641, "upstream_inference_prompt_cost": 0.00134937, "upstream_inference_completions_cost": 0.00011473 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 1348.05, "response_complete": true, "tts_server_stages": [ { "generation_ms": 1840, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1963, "performance": { "device_seconds": 1.6535211882591248, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.6667424146206148, "audio_seconds_per_second": 1.4998295864663314, "batches": [ { "batch_size": 1, "device_seconds": 1.6535211882591248, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.6557160764932632, "first_audio_seconds": 0.2259435577392578, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.001526370644569397 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.48, "quick_reply_tts_ms": 1968.76, "sentence_tts": [ { "sentence": 0, "total_ms": 1968.77, "audio_s": 2.48 } ], "answer_first_audio_ms": 3473.43 }, { "at_ms": 3482.95, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 4244.8, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 4688.01, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 36240 }, { "at_ms": 4740.14, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 4740.28, "text": "Das tut mir sehr leid. Oh, das tut mir so, so sehr leid.\nIch kann mir vorstellen, wie schmerzhaft diese Nachricht ist.", "script": "(contentment, soft, warm) Das tut mir sehr leid. (sadness, soft) Oh, das tut mir so, so sehr leid. (affection, warm) Ich kann mir vorstellen, wie schmerzhaft diese Nachricht ist.", "language": "de", "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das tut mir sehr leid.", "start_s": 0.0, "end_s": 2.48 }, { "emotion": "sadness", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "sadness, soft", "text": "Oh, das tut mir so, so sehr leid.", "start_s": 2.48, "end_s": 6.19 }, { "emotion": "affection", "secondary": null, "blend": 0.0, "intensity": 0.72, "direction": "affection, warm", "text": "Ich kann mir vorstellen, wie schmerzhaft diese Nachricht ist.", "start_s": 6.19, "end_s": 9.25 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "affection.soft", "sentence_index": 2, "sentence_text": "Ich kann mir vorstellen, wie schmerzhaft diese Nachricht ist.", "timing": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "tts_first_audio_ms": 537.82, "first_audio_ms": 851.74, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 1240.67, "llm_first_sentence_ms": 1270.93, "finish_reason": "stop", "usage": { "prompt_tokens": 1363, "completion_tokens": 77, "total_tokens": 1440, "cost": 0.001449459, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0014641, "upstream_inference_prompt_cost": 0.00134937, "upstream_inference_completions_cost": 0.00011473 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 1348.05, "response_complete": true, "tts_server_stages": [ { "generation_ms": 1840, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1963, "performance": { "device_seconds": 1.6535211882591248, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.6667424146206148, "audio_seconds_per_second": 1.4998295864663314, "batches": [ { "batch_size": 1, "device_seconds": 1.6535211882591248, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.6557160764932632, "first_audio_seconds": 0.2259435577392578, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.001526370644569397 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2258, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2437, "performance": { "device_seconds": 1.8303051085472108, "raw_audio_seconds": 3.84, "aggregate_rtf": 0.4766419553508362, "audio_seconds_per_second": 2.098010862816182, "batches": [ { "batch_size": 1, "device_seconds": 1.8303051085472108, "audio_seconds": [ 3.84 ], "model_wall_seconds": 2.0802470222115517, "first_audio_seconds": 0.7658231201171875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.2463180050253868 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.48, "quick_reply_tts_ms": 1968.76, "sentence_tts": [ { "sentence": 0, "total_ms": 1968.77, "audio_s": 2.48 }, { "sentence": 1, "total_ms": 2444.5, "audio_s": 3.71 } ], "answer_first_audio_ms": 3473.43, "sentence_ready_ms": 4729.98, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "affection.soft" } }, { "at_ms": 5919.72, "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "tts_first_audio_ms": 537.82, "first_audio_ms": 851.74, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 1240.67, "llm_first_sentence_ms": 1270.93, "finish_reason": "stop", "usage": { "prompt_tokens": 1363, "completion_tokens": 77, "total_tokens": 1440, "cost": 0.001449459, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0014641, "upstream_inference_prompt_cost": 0.00134937, "upstream_inference_completions_cost": 0.00011473 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 1348.05, "response_complete": true, "tts_server_stages": [ { "generation_ms": 1840, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1963, "performance": { "device_seconds": 1.6535211882591248, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.6667424146206148, "audio_seconds_per_second": 1.4998295864663314, "batches": [ { "batch_size": 1, "device_seconds": 1.6535211882591248, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.6557160764932632, "first_audio_seconds": 0.2259435577392578, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.001526370644569397 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2258, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2437, "performance": { "device_seconds": 1.8303051085472108, "raw_audio_seconds": 3.84, "aggregate_rtf": 0.4766419553508362, "audio_seconds_per_second": 2.098010862816182, "batches": [ { "batch_size": 1, "device_seconds": 1.8303051085472108, "audio_seconds": [ 3.84 ], "model_wall_seconds": 2.0802470222115517, "first_audio_seconds": 0.7658231201171875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.2463180050253868 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.48, "quick_reply_tts_ms": 1968.76, "sentence_tts": [ { "sentence": 0, "total_ms": 1968.77, "audio_s": 2.48 }, { "sentence": 1, "total_ms": 2444.5, "audio_s": 3.71 } ], "answer_first_audio_ms": 3473.43 }, { "at_ms": 5919.9, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 6752.06, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 7263.82, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 17280 }, { "at_ms": 7322.59, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 15120 }, { "at_ms": 7324.26, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 7324.41, "text": "Das tut mir sehr leid. Oh, das tut mir so, so sehr leid.\nIch kann mir vorstellen, wie schmerzhaft diese Nachricht ist.\nEin treuer Freund ist einfach so gegangen.", "script": "(contentment, soft, warm) Das tut mir sehr leid. (sadness, soft) Oh, das tut mir so, so sehr leid. (affection, warm) Ich kann mir vorstellen, wie schmerzhaft diese Nachricht ist. (contemplation, gentle) Ein treuer Freund ist einfach so gegangen.", "language": "de", "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das tut mir sehr leid.", "start_s": 0.0, "end_s": 2.48 }, { "emotion": "sadness", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "sadness, soft", "text": "Oh, das tut mir so, so sehr leid.", "start_s": 2.48, "end_s": 6.19 }, { "emotion": "affection", "secondary": null, "blend": 0.0, "intensity": 0.72, "direction": "affection, warm", "text": "Ich kann mir vorstellen, wie schmerzhaft diese Nachricht ist.", "start_s": 6.19, "end_s": 9.74 }, { "emotion": "contemplation", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "contemplation, gentle", "text": "Ein treuer Freund ist einfach so gegangen.", "start_s": 9.74, "end_s": 12.120000000000001 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "contemplation.soft", "sentence_index": 3, "sentence_text": "Ein treuer Freund ist einfach so gegangen.", "timing": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "tts_first_audio_ms": 537.82, "first_audio_ms": 851.74, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 1240.67, "llm_first_sentence_ms": 1270.93, "finish_reason": "stop", "usage": { "prompt_tokens": 1363, "completion_tokens": 77, "total_tokens": 1440, "cost": 0.001449459, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0014641, "upstream_inference_prompt_cost": 0.00134937, "upstream_inference_completions_cost": 0.00011473 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 1348.05, "response_complete": true, "tts_server_stages": [ { "generation_ms": 1840, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1963, "performance": { "device_seconds": 1.6535211882591248, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.6667424146206148, "audio_seconds_per_second": 1.4998295864663314, "batches": [ { "batch_size": 1, "device_seconds": 1.6535211882591248, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.6557160764932632, "first_audio_seconds": 0.2259435577392578, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.001526370644569397 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2258, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2437, "performance": { "device_seconds": 1.8303051085472108, "raw_audio_seconds": 3.84, "aggregate_rtf": 0.4766419553508362, "audio_seconds_per_second": 2.098010862816182, "batches": [ { "batch_size": 1, "device_seconds": 1.8303051085472108, "audio_seconds": [ 3.84 ], "model_wall_seconds": 2.0802470222115517, "first_audio_seconds": 0.7658231201171875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.2463180050253868 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2407, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2575, "performance": { "device_seconds": 1.8613621401786806, "raw_audio_seconds": 3.92, "aggregate_rtf": 0.4748372806578267, "audio_seconds_per_second": 2.1059845987969332, "batches": [ { "batch_size": 1, "device_seconds": 1.8613621401786806, "audio_seconds": [ 3.92 ], "model_wall_seconds": 2.23168034106493, "first_audio_seconds": 0.7643341064453125, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.36346564441919327 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.48, "quick_reply_tts_ms": 1968.76, "sentence_tts": [ { "sentence": 0, "total_ms": 1968.77, "audio_s": 2.48 }, { "sentence": 1, "total_ms": 2444.5, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 2584.17, "audio_s": 3.55 } ], "answer_first_audio_ms": 3473.43, "sentence_ready_ms": 7314.76, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "contemplation.soft" } }, { "at_ms": 8440.87, "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "tts_first_audio_ms": 537.82, "first_audio_ms": 851.74, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 1240.67, "llm_first_sentence_ms": 1270.93, "finish_reason": "stop", "usage": { "prompt_tokens": 1363, "completion_tokens": 77, "total_tokens": 1440, "cost": 0.001449459, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0014641, "upstream_inference_prompt_cost": 0.00134937, "upstream_inference_completions_cost": 0.00011473 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 1348.05, "response_complete": true, "tts_server_stages": [ { "generation_ms": 1840, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1963, "performance": { "device_seconds": 1.6535211882591248, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.6667424146206148, "audio_seconds_per_second": 1.4998295864663314, "batches": [ { "batch_size": 1, "device_seconds": 1.6535211882591248, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.6557160764932632, "first_audio_seconds": 0.2259435577392578, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.001526370644569397 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2258, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2437, "performance": { "device_seconds": 1.8303051085472108, "raw_audio_seconds": 3.84, "aggregate_rtf": 0.4766419553508362, "audio_seconds_per_second": 2.098010862816182, "batches": [ { "batch_size": 1, "device_seconds": 1.8303051085472108, "audio_seconds": [ 3.84 ], "model_wall_seconds": 2.0802470222115517, "first_audio_seconds": 0.7658231201171875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.2463180050253868 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2407, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2575, "performance": { "device_seconds": 1.8613621401786806, "raw_audio_seconds": 3.92, "aggregate_rtf": 0.4748372806578267, "audio_seconds_per_second": 2.1059845987969332, "batches": [ { "batch_size": 1, "device_seconds": 1.8613621401786806, "audio_seconds": [ 3.92 ], "model_wall_seconds": 2.23168034106493, "first_audio_seconds": 0.7643341064453125, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.36346564441919327 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.48, "quick_reply_tts_ms": 1968.76, "sentence_tts": [ { "sentence": 0, "total_ms": 1968.77, "audio_s": 2.48 }, { "sentence": 1, "total_ms": 2444.5, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 2584.17, "audio_s": 3.55 } ], "answer_first_audio_ms": 3473.43 }, { "at_ms": 8441.04, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 9216.25, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 36480 }, { "at_ms": 9292.19, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 7440 }, { "at_ms": 9295.1, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 9295.29, "text": "Das tut mir sehr leid. Oh, das tut mir so, so sehr leid.\nIch kann mir vorstellen, wie schmerzhaft diese Nachricht ist.\nEin treuer Freund ist einfach so gegangen.\nMöchtest du über ihn erzählen, oder brauchst du gerade nur ein wenig Ruhe?", "script": "(contentment, soft, warm) Das tut mir sehr leid. (sadness, soft) Oh, das tut mir so, so sehr leid. (affection, warm) Ich kann mir vorstellen, wie schmerzhaft diese Nachricht ist. (contemplation, gentle) Ein treuer Freund ist einfach so gegangen. (hope, quiet) Möchtest du über ihn erzählen, oder brauchst du gerade nur ein wenig Ruhe?", "language": "de", "performance": { "sentences": [ { "emotion": "contentment", "secondary": "affection", "blend": 0.25, "intensity": 0.55, "direction": "contentment, soft, warm", "text": "Das tut mir sehr leid.", "start_s": 0.0, "end_s": 2.48 }, { "emotion": "sadness", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "sadness, soft", "text": "Oh, das tut mir so, so sehr leid.", "start_s": 2.48, "end_s": 6.19 }, { "emotion": "affection", "secondary": null, "blend": 0.0, "intensity": 0.72, "direction": "affection, warm", "text": "Ich kann mir vorstellen, wie schmerzhaft diese Nachricht ist.", "start_s": 6.19, "end_s": 9.74 }, { "emotion": "contemplation", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "contemplation, gentle", "text": "Ein treuer Freund ist einfach so gegangen.", "start_s": 9.74, "end_s": 12.17 }, { "emotion": "hope", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "hope, quiet", "text": "Möchtest du über ihn erzählen, oder brauchst du gerade nur ein wenig Ruhe?", "start_s": 12.17, "end_s": 16.59 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "hope.soft", "sentence_index": 4, "sentence_text": "Möchtest du über ihn erzählen, oder brauchst du gerade nur ein wenig Ruhe?", "timing": { "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "tts_first_audio_ms": 537.82, "first_audio_ms": 851.74, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 1240.67, "llm_first_sentence_ms": 1270.93, "finish_reason": "stop", "usage": { "prompt_tokens": 1363, "completion_tokens": 77, "total_tokens": 1440, "cost": 0.001449459, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0014641, "upstream_inference_prompt_cost": 0.00134937, "upstream_inference_completions_cost": 0.00011473 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 1348.05, "response_complete": true, "tts_server_stages": [ { "generation_ms": 1840, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1963, "performance": { "device_seconds": 1.6535211882591248, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.6667424146206148, "audio_seconds_per_second": 1.4998295864663314, "batches": [ { "batch_size": 1, "device_seconds": 1.6535211882591248, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.6557160764932632, "first_audio_seconds": 0.2259435577392578, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.001526370644569397 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2258, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2437, "performance": { "device_seconds": 1.8303051085472108, "raw_audio_seconds": 3.84, "aggregate_rtf": 0.4766419553508362, "audio_seconds_per_second": 2.098010862816182, "batches": [ { "batch_size": 1, "device_seconds": 1.8303051085472108, "audio_seconds": [ 3.84 ], "model_wall_seconds": 2.0802470222115517, "first_audio_seconds": 0.7658231201171875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.2463180050253868 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2407, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2575, "performance": { "device_seconds": 1.8613621401786806, "raw_audio_seconds": 3.92, "aggregate_rtf": 0.4748372806578267, "audio_seconds_per_second": 2.1059845987969332, "batches": [ { "batch_size": 1, "device_seconds": 1.8613621401786806, "audio_seconds": [ 3.92 ], "model_wall_seconds": 2.23168034106493, "first_audio_seconds": 0.7643341064453125, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.36346564441919327 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1755, "alignment_ms": 75, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1961, "performance": { "device_seconds": 1.417962281703949, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.4544750902897272, "audio_seconds_per_second": 2.2003406157254988, "batches": [ { "batch_size": 1, "device_seconds": 1.417962281703949, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.572543017566204, "first_audio_seconds": 0.739330078125, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.15198525041341782 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.48, "quick_reply_tts_ms": 1968.76, "sentence_tts": [ { "sentence": 0, "total_ms": 1968.77, "audio_s": 2.48 }, { "sentence": 1, "total_ms": 2444.5, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 2584.17, "audio_s": 3.55 }, { "sentence": 3, "total_ms": 1969.3, "audio_s": 2.43 } ], "answer_first_audio_ms": 3473.43, "sentence_ready_ms": 9284.98, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "hope.soft" } }, { "at_ms": 10476.59, "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "tts_first_audio_ms": 537.82, "first_audio_ms": 851.74, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 1240.67, "llm_first_sentence_ms": 1270.93, "finish_reason": "stop", "usage": { "prompt_tokens": 1363, "completion_tokens": 77, "total_tokens": 1440, "cost": 0.001449459, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0014641, "upstream_inference_prompt_cost": 0.00134937, "upstream_inference_completions_cost": 0.00011473 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 1348.05, "response_complete": true, "tts_server_stages": [ { "generation_ms": 1840, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1963, "performance": { "device_seconds": 1.6535211882591248, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.6667424146206148, "audio_seconds_per_second": 1.4998295864663314, "batches": [ { "batch_size": 1, "device_seconds": 1.6535211882591248, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.6557160764932632, "first_audio_seconds": 0.2259435577392578, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.001526370644569397 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2258, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2437, "performance": { "device_seconds": 1.8303051085472108, "raw_audio_seconds": 3.84, "aggregate_rtf": 0.4766419553508362, "audio_seconds_per_second": 2.098010862816182, "batches": [ { "batch_size": 1, "device_seconds": 1.8303051085472108, "audio_seconds": [ 3.84 ], "model_wall_seconds": 2.0802470222115517, "first_audio_seconds": 0.7658231201171875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.2463180050253868 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2407, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2575, "performance": { "device_seconds": 1.8613621401786806, "raw_audio_seconds": 3.92, "aggregate_rtf": 0.4748372806578267, "audio_seconds_per_second": 2.1059845987969332, "batches": [ { "batch_size": 1, "device_seconds": 1.8613621401786806, "audio_seconds": [ 3.92 ], "model_wall_seconds": 2.23168034106493, "first_audio_seconds": 0.7643341064453125, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.36346564441919327 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1755, "alignment_ms": 75, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1961, "performance": { "device_seconds": 1.417962281703949, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.4544750902897272, "audio_seconds_per_second": 2.2003406157254988, "batches": [ { "batch_size": 1, "device_seconds": 1.417962281703949, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.572543017566204, "first_audio_seconds": 0.739330078125, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.15198525041341782 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.48, "quick_reply_tts_ms": 1968.76, "sentence_tts": [ { "sentence": 0, "total_ms": 1968.77, "audio_s": 2.48 }, { "sentence": 1, "total_ms": 2444.5, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 2584.17, "audio_s": 3.55 }, { "sentence": 3, "total_ms": 1969.3, "audio_s": 2.43 } ], "answer_first_audio_ms": 3473.43 }, { "at_ms": 10476.78, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 11285.65, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 12157.9, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 12539.01, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 7680 }, { "at_ms": 12599.96, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 17040 }, { "at_ms": 12600.97, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 12601.12, "quick_reply_status": "played", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "quick_reply_generation_ms": 311.92, "quick_reply_model": "nemotron-bolt", "quick_reply_words": 5, "tts_first_audio_ms": 537.82, "first_audio_ms": 851.74, "llm_actual_provider": "Cerebras", "llm_ttft_ms": 1240.67, "llm_first_sentence_ms": 1270.93, "finish_reason": "stop", "usage": { "prompt_tokens": 1363, "completion_tokens": 77, "total_tokens": 1440, "cost": 0.001449459, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0014641, "upstream_inference_prompt_cost": 0.00134937, "upstream_inference_completions_cost": 0.00011473 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "generation_calls": 1, "llm_total_ms": 1348.05, "response_complete": true, "tts_server_stages": [ { "generation_ms": 1840, "alignment_ms": 0, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1963, "performance": { "device_seconds": 1.6535211882591248, "raw_audio_seconds": 2.48, "aggregate_rtf": 0.6667424146206148, "audio_seconds_per_second": 1.4998295864663314, "batches": [ { "batch_size": 1, "device_seconds": 1.6535211882591248, "audio_seconds": [ 2.48 ], "model_wall_seconds": 1.6557160764932632, "first_audio_seconds": 0.2259435577392578, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.001526370644569397 } ], "streaming": true, "chunk_frames": 5, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2258, "alignment_ms": 53, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2437, "performance": { "device_seconds": 1.8303051085472108, "raw_audio_seconds": 3.84, "aggregate_rtf": 0.4766419553508362, "audio_seconds_per_second": 2.098010862816182, "batches": [ { "batch_size": 1, "device_seconds": 1.8303051085472108, "audio_seconds": [ 3.84 ], "model_wall_seconds": 2.0802470222115517, "first_audio_seconds": 0.7658231201171875, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.2463180050253868 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 2407, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2575, "performance": { "device_seconds": 1.8613621401786806, "raw_audio_seconds": 3.92, "aggregate_rtf": 0.4748372806578267, "audio_seconds_per_second": 2.1059845987969332, "batches": [ { "batch_size": 1, "device_seconds": 1.8613621401786806, "audio_seconds": [ 3.92 ], "model_wall_seconds": 2.23168034106493, "first_audio_seconds": 0.7643341064453125, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.36346564441919327 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 1755, "alignment_ms": 75, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 1961, "performance": { "device_seconds": 1.417962281703949, "raw_audio_seconds": 3.12, "aggregate_rtf": 0.4544750902897272, "audio_seconds_per_second": 2.2003406157254988, "batches": [ { "batch_size": 1, "device_seconds": 1.417962281703949, "audio_seconds": [ 3.12 ], "model_wall_seconds": 1.572543017566204, "first_audio_seconds": 0.739330078125, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.15198525041341782 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } }, { "generation_ms": 3124, "alignment_ms": 51, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 3294, "performance": { "device_seconds": 2.431072164058685, "raw_audio_seconds": 5.12, "aggregate_rtf": 0.47481878204271194, "audio_seconds_per_second": 2.1060666465170406, "batches": [ { "batch_size": 1, "device_seconds": 2.431072164058685, "audio_seconds": [ 5.12 ], "model_wall_seconds": 2.9415963888168335, "first_audio_seconds": 0.7714937133789063, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.5047990530729294 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "quick_reply_audio_s": 2.48, "quick_reply_tts_ms": 1968.76, "sentence_tts": [ { "sentence": 0, "total_ms": 1968.77, "audio_s": 2.48 }, { "sentence": 1, "total_ms": 2444.5, "audio_s": 3.71 }, { "sentence": 2, "total_ms": 2584.17, "audio_s": 3.55 }, { "sentence": 3, "total_ms": 1969.3, "audio_s": 2.43 }, { "sentence": 4, "total_ms": 3306.7, "audio_s": 4.83 } ], "answer_first_audio_ms": 3473.43, "total_ms": 12591.87, "audio_s": 17.0, "sentences": 5, "generation_rtf": 0.7407 } ], "first_pcm_ms": 860.82, "first_main_audio_ms": 3482.95, "audio_s": 17.0 }, { "case": "english_loss", "input": "Please speak English. My cat died yesterday.", "history": [], "events": [ { "at_ms": 350.13, "text": "I am so deeply sorry to hear that.", "script": "(sadness, soft) I am so deeply sorry to hear that.", "language": "en", "performance": { "sentences": [ { "emotion": "sadness", "secondary": null, "blend": 0.0, "intensity": 0.55, "direction": "sadness, soft", "text": "I am so deeply sorry to hear that.", "start_s": 0.0, "end_s": 2.72 } ], "source": "qwen/qwen3.8-27b sentence cues + fixed GENERAL catalogue", "timing": "Sentence starts anchored to actual preceding PCM duration; within-sentence estimates" }, "avatar_actions": [], "avatar_action_id": null, "avatar_event": null, "interaction_actions": [], "sources": [], "lookups": [], "speaker": {}, "sentence_role": "answer", "template_id": "sadness.soft", "sentence_index": 0, "sentence_text": "I am so deeply sorry to hear that.", "timing": { "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 302.14, "finish_reason": "stop", "usage": { "prompt_tokens": 1364, "completion_tokens": 16, "total_tokens": 1380, "cost": 0.001360458, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0013742, "upstream_inference_prompt_cost": 0.00135036, "upstream_inference_completions_cost": 2.384e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "llm_first_sentence_ms": 338.18, "generation_calls": 1, "llm_total_ms": 338.2, "response_complete": true, "sentence_ready_ms": 339.28, "director_mode": "bolt", "director_model": "qwen/qwen3.8-27b", "template_id": "sadness.soft" } }, { "at_ms": 1447.16, "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 302.14, "finish_reason": "stop", "usage": { "prompt_tokens": 1364, "completion_tokens": 16, "total_tokens": 1380, "cost": 0.001360458, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0013742, "upstream_inference_prompt_cost": 0.00135036, "upstream_inference_completions_cost": 2.384e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "llm_first_sentence_ms": 338.18, "generation_calls": 1, "llm_total_ms": 338.2, "response_complete": true, "answer_first_audio_ms": 1436.75, "tts_first_audio_ms": 1097.42, "first_audio_ms": 1436.76 }, { "at_ms": 1447.33, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 14390 }, { "at_ms": 2253.01, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 38400 }, { "at_ms": 2712.55, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 15360 }, { "at_ms": 2766.31, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 17040 }, { "at_ms": 2767.05, "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1, "samples": 10 }, { "at_ms": 2767.15, "quick_reply_status": "main_answer_won", "llm_model": "qwen/qwen3.8-27b", "llm_provider": "cerebras/fp16", "llm_backend": "OpenRouter", "llm_actual_provider": "Cerebras", "llm_ttft_ms": 302.14, "finish_reason": "stop", "usage": { "prompt_tokens": 1364, "completion_tokens": 16, "total_tokens": 1380, "cost": 0.001360458, "is_byok": false, "prompt_tokens_details": { "cached_tokens": 1024, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "cost_details": { "upstream_inference_cost": 0.0013742, "upstream_inference_prompt_cost": 0.00135036, "upstream_inference_completions_cost": 2.384e-05 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "llm_first_sentence_ms": 338.18, "generation_calls": 1, "llm_total_ms": 338.2, "response_complete": true, "answer_first_audio_ms": 1436.75, "tts_first_audio_ms": 1097.42, "first_audio_ms": 1436.76, "tts_server_stages": [ { "generation_ms": 2237, "alignment_ms": 47, "scoring_ms": 0, "speaker_preparation_ms": 0, "total_ms": 2407, "performance": { "device_seconds": 1.7806700062751768, "raw_audio_seconds": 3.84, "aggregate_rtf": 0.46371614746749396, "audio_seconds_per_second": 2.1564916500348934, "batches": [ { "batch_size": 1, "device_seconds": 1.7806700062751768, "audio_seconds": [ 3.84 ], "model_wall_seconds": 2.0548474937677383, "first_audio_seconds": 0.7333079223632812, "graph_capture_seconds": 0.0, "live_callback_seconds": 0.2680693194270134 } ], "streaming": true, "chunk_frames": 20, "backend": "fused CFG + local CUDA graphs + compiled global", "includes": "TTS prefill, autoregressive decode and PCM codec; GPU event span including launch gaps", "excludes": "reference encoding, prompt packing, Luna, alignment, scoring, voice conversion, CPU audio copy, MP3, network", "graph_capture_seconds": 0.0, "oom_batch_retries": [] } } ], "sentence_tts": [ { "sentence": 0, "total_ms": 2417.6, "audio_s": 3.55 } ], "total_ms": 2756.95, "audio_s": 3.55, "sentences": 1, "generation_rtf": 0.7766 } ], "first_pcm_ms": 1447.33, "first_main_audio_ms": 1447.33, "audio_s": 3.55 } ] }