{ "schema_version": 1, "created_at": "2026-08-12T10:35:30Z", "database_writes": false, "runs": [ { "combination": "small_vad_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 112.323, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.702, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 1, "window_index": 1, "window_seconds": 19.98, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 1, "window_index": 1, "start_ms": 0, "duration_ms": 19980, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "약불로 잘 쓰여요. 약불로 쓰여요.", "start_ms": 0, "end_ms": 19980 } ], "accepted_output": [ { "text": "약불로 잘 쓰여요. 약불로 쓰여요.", "start_ms": 0, "end_ms": 19980, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 4.573, "rtf": 0.229, "segmentation": "vad" }, { "event": "chunk", "index": 1, "total": 8, "fraction": 0.125, "chunk_index": 1, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 4.573, "rtf": 0.229, "segments": [ { "text": "약불로 잘 쓰여요. 약불로 쓰여요.", "start_ms": 0, "end_ms": 19980, "quality_score": 1.0, "language": "ko" } ] } ], "stderr_tail": "", "window": { "label": "clear_dialogue_1", "start_ms": 25000, "end_ms": 45000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_vad_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 112.323, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.702, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 2, "window_index": 1, "window_seconds": 19.98, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 2, "window_index": 1, "start_ms": 20000, "duration_ms": 19980, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 19980 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 3.155, "rtf": 0.158, "segmentation": "vad" }, { "event": "chunk", "index": 2, "total": 8, "fraction": 0.25, "chunk_index": 2, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 3.155, "rtf": 0.158, "segments": [] } ], "stderr_tail": "", "window": { "label": "clear_dialogue_2", "start_ms": 292000, "end_ms": 312000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_vad_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 112.323, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.702, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 3, "window_index": 1, "window_seconds": 4.44, "max_new_tokens": 30, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 3, "window_index": 1, "start_ms": 40000, "duration_ms": 4440, "speech_ratio": 0.6554, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "다 spent", "start_ms": 0, "end_ms": 4440 } ], "accepted_output": [ { "text": "다 spent", "start_ms": 0, "end_ms": 4440, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.203, "rtf": 0.721, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 3, "window_index": 2, "window_seconds": 4.89, "max_new_tokens": 32, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 3, "window_index": 2, "start_ms": 45550, "duration_ms": 4890, "speech_ratio": 0.7117, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "여기가 좀 더 힘들어요", "start_ms": 0, "end_ms": 4890 } ], "accepted_output": [ { "text": "여기가 좀 더 힘들어요", "start_ms": 0, "end_ms": 4890, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.579, "rtf": 0.732, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 3, "window_index": 3, "window_seconds": 1.71, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 3, "window_index": 3, "start_ms": 54850, "duration_ms": 1710, "speech_ratio": 0.5789, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아,", "start_ms": 0, "end_ms": 1710 } ], "accepted_output": [ { "text": "아,", "start_ms": 0, "end_ms": 1710, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.176, "rtf": 1.857, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 3, "window_index": 4, "window_seconds": 0.9, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 3, "window_index": 4, "start_ms": 58390, "duration_ms": 900, "speech_ratio": 0.6333, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 900 } ], "accepted_output": [ { "text": "아...", "start_ms": 0, "end_ms": 900, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.248, "rtf": 3.609, "segmentation": "vad" }, { "event": "chunk", "index": 3, "total": 8, "fraction": 0.375, "chunk_index": 3, "window_count": 4, "media_seconds": 20.0, "inference_seconds": 13.205, "rtf": 0.66, "segments": [ { "text": "다 spent", "start_ms": 40000, "end_ms": 44440, "quality_score": 1.0, "language": "ko" }, { "text": "여기가 좀 더 힘들어요", "start_ms": 45550, "end_ms": 50440, "quality_score": 1.0, "language": "ko" }, { "text": "아,", "start_ms": 54850, "end_ms": 56560, "quality_score": 1.0, "language": "ko" }, { "text": "아...", "start_ms": 58390, "end_ms": 59290, "quality_score": 1.0, "language": "ko" } ] } ], "stderr_tail": "", "window": { "label": "short_dense_1", "start_ms": 520000, "end_ms": 540000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_vad_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 112.323, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.702, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 4, "window_index": 1, "window_seconds": 0.48, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 1, "start_ms": 60000, "duration_ms": 480, "speech_ratio": 0.5, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "MBC News.", "start_ms": 0, "end_ms": 480 } ], "accepted_output": [ { "text": "MBC News.", "start_ms": 0, "end_ms": 480, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.471, "rtf": 7.231, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 4, "window_index": 2, "window_seconds": 1.62, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 2, "start_ms": 64950, "duration_ms": 1620, "speech_ratio": 0.5, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 1620 } ], "accepted_output": [ { "text": "아...", "start_ms": 0, "end_ms": 1620, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.156, "rtf": 1.948, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 4, "window_index": 3, "window_seconds": 0.72, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 3, "start_ms": 68010, "duration_ms": 720, "speech_ratio": 0.3333, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "맛있어.", "start_ms": 0, "end_ms": 720 } ], "accepted_output": [ { "text": "맛있어.", "start_ms": 0, "end_ms": 720, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.201, "rtf": 4.446, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 4, "window_index": 4, "window_seconds": 0.81, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 4, "start_ms": 70920, "duration_ms": 810, "speech_ratio": 0.7407, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "MBC 뉴스 김지경입니다.", "start_ms": 0, "end_ms": 810 } ], "accepted_output": [ { "text": "MBC 뉴스 김지경입니다.", "start_ms": 0, "end_ms": 810, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.403, "rtf": 4.202, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 4, "window_index": 5, "window_seconds": 6.69, "max_new_tokens": 39, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 5, "start_ms": 73020, "duration_ms": 6690, "speech_ratio": 0.6009, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 6690 } ], "accepted_output": [ { "text": "아...", "start_ms": 0, "end_ms": 6690, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 2.998, "rtf": 0.448, "segmentation": "vad" }, { "event": "chunk", "index": 4, "total": 8, "fraction": 0.5, "chunk_index": 4, "window_count": 5, "media_seconds": 20.0, "inference_seconds": 16.228, "rtf": 0.811, "segments": [ { "text": "MBC News.", "start_ms": 60000, "end_ms": 60480, "quality_score": 1.0, "language": "ko" }, { "text": "아...", "start_ms": 64950, "end_ms": 66570, "quality_score": 1.0, "language": "ko" }, { "text": "맛있어.", "start_ms": 68010, "end_ms": 68730, "quality_score": 1.0, "language": "ko" }, { "text": "MBC 뉴스 김지경입니다.", "start_ms": 70920, "end_ms": 71730, "quality_score": 1.0, "language": "ko" }, { "text": "아...", "start_ms": 73020, "end_ms": 79710, "quality_score": 1.0, "language": "ko" } ] } ], "stderr_tail": "", "window": { "label": "short_dense_2", "start_ms": 545000, "end_ms": 565000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_vad_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 112.323, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.702, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 5, "window_index": 1, "window_seconds": 1.62, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 5, "window_index": 1, "start_ms": 81950, "duration_ms": 1620, "speech_ratio": 0.5, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 1620 } ], "accepted_output": [ { "text": "아...", "start_ms": 0, "end_ms": 1620, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.186, "rtf": 1.967, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 5, "window_index": 2, "window_seconds": 0.72, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 5, "window_index": 2, "start_ms": 85010, "duration_ms": 720, "speech_ratio": 0.3333, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "맛있어.", "start_ms": 0, "end_ms": 720 } ], "accepted_output": [ { "text": "맛있어.", "start_ms": 0, "end_ms": 720, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.246, "rtf": 4.508, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 5, "window_index": 3, "window_seconds": 0.81, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 5, "window_index": 3, "start_ms": 87920, "duration_ms": 810, "speech_ratio": 0.7407, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "MBC 뉴스 김지경입니다.", "start_ms": 0, "end_ms": 810 } ], "accepted_output": [ { "text": "MBC 뉴스 김지경입니다.", "start_ms": 0, "end_ms": 810, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.675, "rtf": 4.537, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 5, "window_index": 4, "window_seconds": 7.71, "max_new_tokens": 43, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 5, "window_index": 4, "start_ms": 90020, "duration_ms": 7710, "speech_ratio": 0.6265, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 7710 } ], "accepted_output": [ { "text": "아...", "start_ms": 0, "end_ms": 7710, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.109, "rtf": 0.403, "segmentation": "vad" }, { "event": "chunk", "index": 5, "total": 8, "fraction": 0.625, "chunk_index": 5, "window_count": 4, "media_seconds": 20.0, "inference_seconds": 13.215, "rtf": 0.661, "segments": [ { "text": "아...", "start_ms": 81950, "end_ms": 83570, "quality_score": 1.0, "language": "ko" }, { "text": "맛있어.", "start_ms": 85010, "end_ms": 85730, "quality_score": 1.0, "language": "ko" }, { "text": "MBC 뉴스 김지경입니다.", "start_ms": 87920, "end_ms": 88730, "quality_score": 1.0, "language": "ko" }, { "text": "아...", "start_ms": 90020, "end_ms": 97730, "quality_score": 1.0, "language": "ko" } ] } ], "stderr_tail": "", "window": { "label": "suspected_hallucination_1", "start_ms": 548000, "end_ms": 568000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_vad_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 112.323, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.702, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 6, "window_index": 1, "window_seconds": 19.98, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 6, "window_index": 1, "start_ms": 100000, "duration_ms": 19980, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 19980 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 3.214, "rtf": 0.161, "segmentation": "vad" }, { "event": "chunk", "index": 6, "total": 8, "fraction": 0.75, "chunk_index": 6, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 3.214, "rtf": 0.161, "segments": [] } ], "stderr_tail": "", "window": { "label": "suspected_hallucination_2", "start_ms": 333000, "end_ms": 353000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_vad_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 112.323, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.702, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 7, "window_index": 1, "window_seconds": 19.98, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 7, "window_index": 1, "start_ms": 120000, "duration_ms": 19980, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "다시!", "start_ms": 0, "end_ms": 19980 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 3.135, "rtf": 0.157, "segmentation": "vad" }, { "event": "chunk", "index": 7, "total": 8, "fraction": 0.875, "chunk_index": 7, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 3.134, "rtf": 0.157, "segments": [] } ], "stderr_tail": "", "window": { "label": "vad_miss_1", "start_ms": 200000, "end_ms": 220000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_vad_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 112.323, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.702, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 8, "window_index": 1, "window_seconds": 19.98, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 8, "window_index": 1, "start_ms": 140000, "duration_ms": 19980, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "오, 오, 오, 오, 오, 오, 오, 오, 오, 오, 오, 오, 오, 오, 오,", "start_ms": 0, "end_ms": 19980 } ], "accepted_output": [], "quality_score": 0.0, "quality_flags": [ "repeated_characters", "repetitive_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 5.428, "rtf": 0.272, "segmentation": "vad" }, { "event": "chunk", "index": 8, "total": 8, "fraction": 1.0, "chunk_index": 8, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 5.428, "rtf": 0.271, "segments": [] }, { "event": "complete", "detected_language": "ko", "quality_score": 0.503, "rejected_segments": 4, "quality_flags": [ "too_little_text", "repeated_characters", "repetitive_text", "whole_repeated_phrase", "repeated_phrase:아:6" ], "inference_diagnostics": { "requested_device": "GPU", "actual_device": "GPU.0", "primary_backend": "direct_generate", "model_variant": "small", "num_beams": 1, "segmentation": "vad", "quality_score_semantics": "rule_check", "language_candidates": [ "ko", "ko", "ko" ], "selected_language": "ko", "fallback_scope": null, "fallback_reason": null } } ], "stderr_tail": "", "window": { "label": "vad_miss_2", "start_ms": 400000, "end_ms": 420000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "continuous", "elapsed_seconds": 90.305, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.564, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 1, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 1, "window_index": 1, "start_ms": 0, "duration_ms": 20000, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "약불로 잘 쓰여요. 약불로 쓰여요.", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [ { "text": "약불로 잘 쓰여요. 약불로 쓰여요.", "start_ms": 0, "end_ms": 20000, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 6.742, "rtf": 0.337, "segmentation": "continuous" }, { "event": "chunk", "index": 1, "total": 8, "fraction": 0.125, "chunk_index": 1, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 6.742, "rtf": 0.337, "segments": [ { "text": "약불로 잘 쓰여요. 약불로 쓰여요.", "start_ms": 0, "end_ms": 20000, "quality_score": 1.0, "language": "ko" } ] } ], "stderr_tail": "", "window": { "label": "clear_dialogue_1", "start_ms": 25000, "end_ms": 45000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "continuous", "elapsed_seconds": 90.305, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.564, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 2, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 2, "window_index": 1, "start_ms": 20000, "duration_ms": 20000, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 3.732, "rtf": 0.187, "segmentation": "continuous" }, { "event": "chunk", "index": 2, "total": 8, "fraction": 0.25, "chunk_index": 2, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 3.732, "rtf": 0.187, "segments": [] } ], "stderr_tail": "", "window": { "label": "clear_dialogue_2", "start_ms": 292000, "end_ms": 312000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "continuous", "elapsed_seconds": 90.305, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.564, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 3, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 3, "window_index": 1, "start_ms": 40000, "duration_ms": 20000, "speech_ratio": 0.533, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "여기不用 use for tonight you are very", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [ { "text": "여기不用 use for tonight you are very", "start_ms": 0, "end_ms": 20000, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.893, "rtf": 0.195, "segmentation": "continuous" }, { "event": "chunk", "index": 3, "total": 8, "fraction": 0.375, "chunk_index": 3, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 3.893, "rtf": 0.195, "segments": [ { "text": "여기不用 use for tonight you are very", "start_ms": 40000, "end_ms": 60000, "quality_score": 1.0, "language": "ko" } ] } ], "stderr_tail": "", "window": { "label": "short_dense_1", "start_ms": 520000, "end_ms": 540000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "continuous", "elapsed_seconds": 90.305, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.564, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 4, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 1, "start_ms": 60000, "duration_ms": 20000, "speech_ratio": 0.3979, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [ { "text": "아...", "start_ms": 0, "end_ms": 20000, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 3.09, "rtf": 0.154, "segmentation": "continuous" }, { "event": "chunk", "index": 4, "total": 8, "fraction": 0.5, "chunk_index": 4, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 3.089, "rtf": 0.154, "segments": [ { "text": "아...", "start_ms": 60000, "end_ms": 80000, "quality_score": 1.0, "language": "ko" } ] } ], "stderr_tail": "", "window": { "label": "short_dense_2", "start_ms": 545000, "end_ms": 565000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "continuous", "elapsed_seconds": 90.305, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.564, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 5, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 5, "window_index": 1, "start_ms": 80000, "duration_ms": 20000, "speech_ratio": 0.4264, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 3.061, "rtf": 0.153, "segmentation": "continuous" }, { "event": "chunk", "index": 5, "total": 8, "fraction": 0.625, "chunk_index": 5, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 3.061, "rtf": 0.153, "segments": [] } ], "stderr_tail": "", "window": { "label": "suspected_hallucination_1", "start_ms": 548000, "end_ms": 568000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "continuous", "elapsed_seconds": 90.305, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.564, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 6, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 6, "window_index": 1, "start_ms": 100000, "duration_ms": 20000, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아...", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 3.295, "rtf": 0.165, "segmentation": "continuous" }, { "event": "chunk", "index": 6, "total": 8, "fraction": 0.75, "chunk_index": 6, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 3.295, "rtf": 0.165, "segments": [] } ], "stderr_tail": "", "window": { "label": "suspected_hallucination_2", "start_ms": 333000, "end_ms": 353000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "continuous", "elapsed_seconds": 90.305, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.564, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 7, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 7, "window_index": 1, "start_ms": 120000, "duration_ms": 20000, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "다시.", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 3.158, "rtf": 0.158, "segmentation": "continuous" }, { "event": "chunk", "index": 7, "total": 8, "fraction": 0.875, "chunk_index": 7, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 3.158, "rtf": 0.158, "segments": [] } ], "stderr_tail": "", "window": { "label": "vad_miss_1", "start_ms": 200000, "end_ms": 220000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_gpu_1beam", "model": "small", "device": "GPU", "beam": 1, "segmentation": "continuous", "elapsed_seconds": 90.305, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 0.564, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 8, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 8, "window_index": 1, "start_ms": 140000, "duration_ms": 20000, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오오", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [], "quality_score": 0.0, "quality_flags": [ "repeated_characters", "low_character_diversity", "repetitive_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 11.171, "rtf": 0.559, "segmentation": "continuous" }, { "event": "chunk", "index": 8, "total": 8, "fraction": 1.0, "chunk_index": 8, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 11.171, "rtf": 0.559, "segments": [] }, { "event": "complete", "detected_language": "ko", "quality_score": 0.6, "rejected_segments": 5, "quality_flags": [ "too_little_text", "repeated_characters", "low_character_diversity", "repetitive_text" ], "inference_diagnostics": { "requested_device": "GPU", "actual_device": "GPU.0", "primary_backend": "direct_generate", "model_variant": "small", "num_beams": 1, "segmentation": "continuous", "quality_score_semantics": "rule_check", "language_candidates": [ "ko", "ko", "ko" ], "selected_language": "ko", "fallback_scope": null, "fallback_reason": null } } ], "stderr_tail": "", "window": { "label": "vad_miss_2", "start_ms": 400000, "end_ms": 420000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_cpu_5beam", "model": "small", "device": "CPU", "beam": 5, "segmentation": "continuous", "elapsed_seconds": 170.67, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.067, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 1, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "CPU", "num_beams": 5, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 1, "window_index": 1, "start_ms": 0, "duration_ms": 20000, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "그냥 빠졌어요. 그냥 빠졌어요. 그냥 빠졌어요.", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [ { "text": "그냥 빠졌어요. 그냥 빠졌어요. 그냥 빠졌어요.", "start_ms": 0, "end_ms": 20000, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "CPU", "beam": 5, "elapsed_seconds": 15.269, "rtf": 0.763, "segmentation": "continuous" }, { "event": "chunk", "index": 1, "total": 8, "fraction": 0.125, "chunk_index": 1, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 15.268, "rtf": 0.763, "segments": [ { "text": "그냥 빠졌어요. 그냥 빠졌어요. 그냥 빠졌어요.", "start_ms": 0, "end_ms": 20000, "quality_score": 1.0, "language": "ko" } ] } ], "stderr_tail": "", "window": { "label": "clear_dialogue_1", "start_ms": 25000, "end_ms": 45000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_cpu_5beam", "model": "small", "device": "CPU", "beam": 5, "segmentation": "continuous", "elapsed_seconds": 170.67, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.067, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 2, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "CPU", "num_beams": 5, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 2, "window_index": 1, "start_ms": 20000, "duration_ms": 20000, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "오오오!", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "CPU", "beam": 5, "elapsed_seconds": 12.429, "rtf": 0.621, "segmentation": "continuous" }, { "event": "chunk", "index": 2, "total": 8, "fraction": 0.25, "chunk_index": 2, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 12.429, "rtf": 0.621, "segments": [] } ], "stderr_tail": "", "window": { "label": "clear_dialogue_2", "start_ms": 292000, "end_ms": 312000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_cpu_5beam", "model": "small", "device": "CPU", "beam": 5, "segmentation": "continuous", "elapsed_seconds": 170.67, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.067, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 3, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "CPU", "num_beams": 5, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 3, "window_index": 1, "start_ms": 40000, "duration_ms": 20000, "speech_ratio": 0.533, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "여기 사용하지 마세요. 여기 사용하지 마세요.", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [ { "text": "여기 사용하지 마세요. 여기 사용하지 마세요.", "start_ms": 0, "end_ms": 20000, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "CPU", "beam": 5, "elapsed_seconds": 13.257, "rtf": 0.663, "segmentation": "continuous" }, { "event": "chunk", "index": 3, "total": 8, "fraction": 0.375, "chunk_index": 3, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 13.257, "rtf": 0.663, "segments": [ { "text": "여기 사용하지 마세요. 여기 사용하지 마세요.", "start_ms": 40000, "end_ms": 60000, "quality_score": 1.0, "language": "ko" } ] } ], "stderr_tail": "", "window": { "label": "short_dense_1", "start_ms": 520000, "end_ms": 540000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_cpu_5beam", "model": "small", "device": "CPU", "beam": 5, "segmentation": "continuous", "elapsed_seconds": 170.67, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.067, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 4, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "CPU", "num_beams": 5, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 1, "start_ms": 60000, "duration_ms": 20000, "speech_ratio": 0.3979, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "앗!", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [ { "text": "앗!", "start_ms": 0, "end_ms": 20000, "quality_score": 1.0, "language": "ko" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "CPU", "beam": 5, "elapsed_seconds": 12.237, "rtf": 0.612, "segmentation": "continuous" }, { "event": "chunk", "index": 4, "total": 8, "fraction": 0.5, "chunk_index": 4, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 12.237, "rtf": 0.612, "segments": [ { "text": "앗!", "start_ms": 60000, "end_ms": 80000, "quality_score": 1.0, "language": "ko" } ] } ], "stderr_tail": "", "window": { "label": "short_dense_2", "start_ms": 545000, "end_ms": 565000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_cpu_5beam", "model": "small", "device": "CPU", "beam": 5, "segmentation": "continuous", "elapsed_seconds": 170.67, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.067, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 5, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "CPU", "num_beams": 5, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 5, "window_index": 1, "start_ms": 80000, "duration_ms": 20000, "speech_ratio": 0.4264, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아... 아...", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "CPU", "beam": 5, "elapsed_seconds": 11.683, "rtf": 0.584, "segmentation": "continuous" }, { "event": "chunk", "index": 5, "total": 8, "fraction": 0.625, "chunk_index": 5, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 11.683, "rtf": 0.584, "segments": [] } ], "stderr_tail": "", "window": { "label": "suspected_hallucination_1", "start_ms": 548000, "end_ms": 568000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_cpu_5beam", "model": "small", "device": "CPU", "beam": 5, "segmentation": "continuous", "elapsed_seconds": 170.67, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.067, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 6, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "CPU", "num_beams": 5, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 6, "window_index": 1, "start_ms": 100000, "duration_ms": 20000, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "오오오오오오오오오", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "CPU", "beam": 5, "elapsed_seconds": 13.707, "rtf": 0.685, "segmentation": "continuous" }, { "event": "chunk", "index": 6, "total": 8, "fraction": 0.75, "chunk_index": 6, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 13.707, "rtf": 0.685, "segments": [] } ], "stderr_tail": "", "window": { "label": "suspected_hallucination_2", "start_ms": 333000, "end_ms": 353000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_cpu_5beam", "model": "small", "device": "CPU", "beam": 5, "segmentation": "continuous", "elapsed_seconds": 170.67, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.067, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 7, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "CPU", "num_beams": 5, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 7, "window_index": 1, "start_ms": 120000, "duration_ms": 20000, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "자, 자식.", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "CPU", "beam": 5, "elapsed_seconds": 12.927, "rtf": 0.646, "segmentation": "continuous" }, { "event": "chunk", "index": 7, "total": 8, "fraction": 0.875, "chunk_index": 7, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 12.927, "rtf": 0.646, "segments": [] } ], "stderr_tail": "", "window": { "label": "vad_miss_1", "start_ms": 200000, "end_ms": 220000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "small_context_cpu_5beam", "model": "small", "device": "CPU", "beam": 5, "segmentation": "continuous", "elapsed_seconds": 170.67, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.067, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 8, "window_index": 1, "window_seconds": 20.0, "max_new_tokens": 92, "device": "CPU", "num_beams": 5, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 8, "window_index": 1, "start_ms": 140000, "duration_ms": 20000, "speech_ratio": 1.0, "language_candidates": [ "ko", "ko", "ko" ], "forced_language": "ko", "raw_output": [ { "text": "아 차아 차아 차아 차아", "start_ms": 0, "end_ms": 20000 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "CPU", "beam": 5, "elapsed_seconds": 13.152, "rtf": 0.658, "segmentation": "continuous" }, { "event": "chunk", "index": 8, "total": 8, "fraction": 1.0, "chunk_index": 8, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 13.152, "rtf": 0.658, "segments": [] }, { "event": "complete", "detected_language": "ko", "quality_score": 0.656, "rejected_segments": 5, "quality_flags": [ "too_little_text" ], "inference_diagnostics": { "requested_device": "CPU", "actual_device": "CPU", "primary_backend": "direct_generate", "model_variant": "small", "num_beams": 5, "segmentation": "continuous", "quality_score_semantics": "rule_check", "language_candidates": [ "ko", "ko", "ko" ], "selected_language": "ko", "fallback_scope": null, "fallback_reason": null } } ], "stderr_tail": "", "window": { "label": "vad_miss_2", "start_ms": 400000, "end_ms": 420000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "medium_current_baseline", "model": "medium", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 305.438, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.909, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 1, "window_index": 1, "window_seconds": 19.98, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 1, "window_index": 1, "start_ms": 0, "duration_ms": 19980, "speech_ratio": 1.0, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "你不要太着急啊", "start_ms": 0, "end_ms": 19980 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 10.903, "rtf": 0.546, "segmentation": "vad" }, { "event": "chunk", "index": 1, "total": 8, "fraction": 0.125, "chunk_index": 1, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 10.903, "rtf": 0.545, "segments": [] } ], "stderr_tail": "", "window": { "label": "clear_dialogue_1", "start_ms": 25000, "end_ms": 45000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "medium_current_baseline", "model": "medium", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 305.438, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.909, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 2, "window_index": 1, "window_seconds": 19.98, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 2, "window_index": 1, "start_ms": 20000, "duration_ms": 19980, "speech_ratio": 1.0, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦哦", "start_ms": 0, "end_ms": 19980 } ], "accepted_output": [], "quality_score": 0.0, "quality_flags": [ "repeated_characters", "low_character_diversity", "repetitive_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 23.287, "rtf": 1.165, "segmentation": "vad" }, { "event": "chunk", "index": 2, "total": 8, "fraction": 0.25, "chunk_index": 2, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 23.286, "rtf": 1.164, "segments": [] } ], "stderr_tail": "", "window": { "label": "clear_dialogue_2", "start_ms": 292000, "end_ms": 312000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "medium_current_baseline", "model": "medium", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 305.438, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.909, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 3, "window_index": 1, "window_seconds": 4.44, "max_new_tokens": 30, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 3, "window_index": 1, "start_ms": 40000, "duration_ms": 4440, "speech_ratio": 0.6554, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "我去拿", "start_ms": 0, "end_ms": 4440 } ], "accepted_output": [ { "text": "我去拿", "start_ms": 0, "end_ms": 4440, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 10.667, "rtf": 2.403, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 3, "window_index": 2, "window_seconds": 4.89, "max_new_tokens": 32, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 3, "window_index": 2, "start_ms": 45550, "duration_ms": 4890, "speech_ratio": 0.7117, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "你不用用縮鏡管你就很靈活", "start_ms": 0, "end_ms": 4890 } ], "accepted_output": [ { "text": "你不用用縮鏡管你就很靈活", "start_ms": 0, "end_ms": 4890, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 11.829, "rtf": 2.419, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 3, "window_index": 3, "window_seconds": 1.71, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 3, "window_index": 3, "start_ms": 54850, "duration_ms": 1710, "speech_ratio": 0.5789, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "好", "start_ms": 0, "end_ms": 1710 } ], "accepted_output": [ { "text": "好", "start_ms": 0, "end_ms": 1710, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 10.633, "rtf": 6.218, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 3, "window_index": 4, "window_seconds": 0.9, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 3, "window_index": 4, "start_ms": 58390, "duration_ms": 900, "speech_ratio": 0.6333, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "好", "start_ms": 0, "end_ms": 900 } ], "accepted_output": [ { "text": "好", "start_ms": 0, "end_ms": 900, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 10.392, "rtf": 11.547, "segmentation": "vad" }, { "event": "chunk", "index": 3, "total": 8, "fraction": 0.375, "chunk_index": 3, "window_count": 4, "media_seconds": 20.0, "inference_seconds": 43.521, "rtf": 2.176, "segments": [ { "text": "我去拿", "start_ms": 40000, "end_ms": 44440, "quality_score": 1.0, "language": "zh" }, { "text": "你不用用縮鏡管你就很靈活", "start_ms": 45550, "end_ms": 50440, "quality_score": 1.0, "language": "zh" }, { "text": "好", "start_ms": 54850, "end_ms": 56560, "quality_score": 1.0, "language": "zh" }, { "text": "好", "start_ms": 58390, "end_ms": 59290, "quality_score": 1.0, "language": "zh" } ] } ], "stderr_tail": "", "window": { "label": "short_dense_1", "start_ms": 520000, "end_ms": 540000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "medium_current_baseline", "model": "medium", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 305.438, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.909, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 4, "window_index": 1, "window_seconds": 0.48, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 1, "start_ms": 60000, "duration_ms": 480, "speech_ratio": 0.5, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "好", "start_ms": 0, "end_ms": 480 } ], "accepted_output": [ { "text": "好", "start_ms": 0, "end_ms": 480, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 10.292, "rtf": 21.442, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 4, "window_index": 2, "window_seconds": 1.62, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 2, "start_ms": 64950, "duration_ms": 1620, "speech_ratio": 0.5, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "啊", "start_ms": 0, "end_ms": 1620 } ], "accepted_output": [ { "text": "啊", "start_ms": 0, "end_ms": 1620, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 10.286, "rtf": 6.349, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 4, "window_index": 3, "window_seconds": 0.72, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 3, "start_ms": 68010, "duration_ms": 720, "speech_ratio": 0.3333, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "好吃", "start_ms": 0, "end_ms": 720 } ], "accepted_output": [ { "text": "好吃", "start_ms": 0, "end_ms": 720, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 10.4, "rtf": 14.444, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 4, "window_index": 4, "window_seconds": 0.81, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 4, "start_ms": 70920, "duration_ms": 810, "speech_ratio": 0.7407, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "好", "start_ms": 0, "end_ms": 810 } ], "accepted_output": [ { "text": "好", "start_ms": 0, "end_ms": 810, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 12.08, "rtf": 14.913, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 4, "window_index": 5, "window_seconds": 6.69, "max_new_tokens": 39, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 4, "window_index": 5, "start_ms": 73020, "duration_ms": 6690, "speech_ratio": 0.6009, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "啊", "start_ms": 0, "end_ms": 6690 } ], "accepted_output": [ { "text": "啊", "start_ms": 0, "end_ms": 6690, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 16.952, "rtf": 2.534, "segmentation": "vad" }, { "event": "chunk", "index": 4, "total": 8, "fraction": 0.5, "chunk_index": 4, "window_count": 5, "media_seconds": 20.0, "inference_seconds": 60.009, "rtf": 3.0, "segments": [ { "text": "好", "start_ms": 60000, "end_ms": 60480, "quality_score": 1.0, "language": "zh" }, { "text": "啊", "start_ms": 64950, "end_ms": 66570, "quality_score": 1.0, "language": "zh" }, { "text": "好吃", "start_ms": 68010, "end_ms": 68730, "quality_score": 1.0, "language": "zh" }, { "text": "好", "start_ms": 70920, "end_ms": 71730, "quality_score": 1.0, "language": "zh" }, { "text": "啊", "start_ms": 73020, "end_ms": 79710, "quality_score": 1.0, "language": "zh" } ] } ], "stderr_tail": "", "window": { "label": "short_dense_2", "start_ms": 545000, "end_ms": 565000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "medium_current_baseline", "model": "medium", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 305.438, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.909, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 5, "window_index": 1, "window_seconds": 1.62, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 5, "window_index": 1, "start_ms": 81950, "duration_ms": 1620, "speech_ratio": 0.5, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "啊", "start_ms": 0, "end_ms": 1620 } ], "accepted_output": [ { "text": "啊", "start_ms": 0, "end_ms": 1620, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 12.188, "rtf": 7.523, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 5, "window_index": 2, "window_seconds": 0.72, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 5, "window_index": 2, "start_ms": 85010, "duration_ms": 720, "speech_ratio": 0.3333, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "好吃", "start_ms": 0, "end_ms": 720 } ], "accepted_output": [ { "text": "好吃", "start_ms": 0, "end_ms": 720, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 13.643, "rtf": 18.949, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 5, "window_index": 3, "window_seconds": 0.81, "max_new_tokens": 24, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 5, "window_index": 3, "start_ms": 87920, "duration_ms": 810, "speech_ratio": 0.7407, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "好", "start_ms": 0, "end_ms": 810 } ], "accepted_output": [ { "text": "好", "start_ms": 0, "end_ms": 810, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 13.404, "rtf": 16.548, "segmentation": "vad" }, { "event": "decode_start", "chunk_index": 5, "window_index": 4, "window_seconds": 7.71, "max_new_tokens": 43, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 5, "window_index": 4, "start_ms": 90020, "duration_ms": 7710, "speech_ratio": 0.6265, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "啊", "start_ms": 0, "end_ms": 7710 } ], "accepted_output": [ { "text": "啊", "start_ms": 0, "end_ms": 7710, "quality_score": 1.0, "language": "zh" } ], "quality_score": 1.0, "quality_flags": [], "rejected_segments": 0, "device": "GPU", "beam": 1, "elapsed_seconds": 12.712, "rtf": 1.649, "segmentation": "vad" }, { "event": "chunk", "index": 5, "total": 8, "fraction": 0.625, "chunk_index": 5, "window_count": 4, "media_seconds": 20.0, "inference_seconds": 51.946, "rtf": 2.597, "segments": [ { "text": "啊", "start_ms": 81950, "end_ms": 83570, "quality_score": 1.0, "language": "zh" }, { "text": "好吃", "start_ms": 85010, "end_ms": 85730, "quality_score": 1.0, "language": "zh" }, { "text": "好", "start_ms": 87920, "end_ms": 88730, "quality_score": 1.0, "language": "zh" }, { "text": "啊", "start_ms": 90020, "end_ms": 97730, "quality_score": 1.0, "language": "zh" } ] } ], "stderr_tail": "", "window": { "label": "suspected_hallucination_1", "start_ms": 548000, "end_ms": 568000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "medium_current_baseline", "model": "medium", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 305.438, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.909, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 6, "window_index": 1, "window_seconds": 19.98, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 6, "window_index": 1, "start_ms": 100000, "duration_ms": 19980, "speech_ratio": 1.0, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯嗯", "start_ms": 0, "end_ms": 19980 } ], "accepted_output": [], "quality_score": 0.0, "quality_flags": [ "repeated_characters", "low_character_diversity", "repetitive_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 29.076, "rtf": 1.455, "segmentation": "vad" }, { "event": "chunk", "index": 6, "total": 8, "fraction": 0.75, "chunk_index": 6, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 29.074, "rtf": 1.454, "segments": [] } ], "stderr_tail": "", "window": { "label": "suspected_hallucination_2", "start_ms": 333000, "end_ms": 353000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "medium_current_baseline", "model": "medium", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 305.438, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.909, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 7, "window_index": 1, "window_seconds": 19.98, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 7, "window_index": 1, "start_ms": 120000, "duration_ms": 19980, "speech_ratio": 1.0, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "謝謝", "start_ms": 0, "end_ms": 19980 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 14.881, "rtf": 0.745, "segmentation": "vad" }, { "event": "chunk", "index": 7, "total": 8, "fraction": 0.875, "chunk_index": 7, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 14.88, "rtf": 0.744, "segments": [] } ], "stderr_tail": "", "window": { "label": "vad_miss_1", "start_ms": 200000, "end_ms": 220000 }, "media": "g4ww_FBcUUc7YTPo.mp4" }, { "combination": "medium_current_baseline", "model": "medium", "device": "GPU", "beam": 1, "segmentation": "vad", "elapsed_seconds": 305.438, "window_seconds": 20.0, "batch_window_count": 8, "batch_rtf": 1.909, "exit_code": 0, "events": [ { "event": "decode_start", "chunk_index": 8, "window_index": 1, "window_seconds": 19.98, "max_new_tokens": 92, "device": "GPU", "num_beams": 1, "batch_size": 1, "cache": true }, { "event": "window_result", "chunk_index": 8, "window_index": 1, "start_ms": 140000, "duration_ms": 19980, "speech_ratio": 1.0, "language_candidates": [ "en", "ja", "ko" ], "forced_language": "zh", "raw_output": [ { "text": "好", "start_ms": 0, "end_ms": 19980 } ], "accepted_output": [], "quality_score": 0.45, "quality_flags": [ "too_little_text" ], "rejected_segments": 1, "device": "GPU", "beam": 1, "elapsed_seconds": 14.137, "rtf": 0.708, "segmentation": "vad" }, { "event": "chunk", "index": 8, "total": 8, "fraction": 1.0, "chunk_index": 8, "window_count": 1, "media_seconds": 20.0, "inference_seconds": 14.137, "rtf": 0.707, "segments": [] }, { "event": "complete", "detected_language": "zh", "quality_score": 0.097, "rejected_segments": 5, "quality_flags": [ "too_little_text", "repeated_characters", "low_character_diversity", "repetitive_text", "whole_repeated_phrase", "whole_short_segment_dominance", "repeated_phrase:好:5", "repeated_phrase:啊:4" ], "inference_diagnostics": { "requested_device": "GPU", "actual_device": "GPU.0", "primary_backend": "direct_generate", "model_variant": "medium", "num_beams": 1, "segmentation": "vad", "quality_score_semantics": "rule_check", "language_candidates": [ "en", "ja", "ko" ], "selected_language": "zh", "fallback_scope": null, "fallback_reason": null } } ], "stderr_tail": "", "window": { "label": "vad_miss_2", "start_ms": 400000, "end_ms": 420000 }, "media": "g4ww_FBcUUc7YTPo.mp4" } ] }