{ "scope": "Four real webcam-microphone questions, local recognition and accelerator LLM plus USB playback", "reference_note": "Round1 prescribed wording confirmed by user; round2 compared to prescribed wording. User reported first reply incorrect and second useful; USB playback returned successfully. Audibility was not separately confirmed.", "token_limit": 32, "timing_note": "Question end estimated from amplitude. ALSA submit/complete is not acoustic timing. Setup before cue excluded. Outputs stopped at 32-token limit, not EOS.", "exact_matches": 1, "cases": [ { "case": "round1_paris", "expected_question": "パリはどこにありますか", "recognized": "あるのはどこにありますか？", "exact_normalized_match": false, "character_error_rate": 0.2727272727272727, "mic_gain_percent": 35, "rms": 394, "peak": 3807, "clipped_fraction": 0.0, "recorded_audio_seconds": 3.44, "capture": { "interaction_start_monotonic": 10958.818081874, "cue_monotonic": 10960.506521193, "capture_finished_monotonic": 10964.048016585, "capture_seconds": 3.5414953920007974, "first_voice_offset_seconds": 0.94, "last_voice_offset_seconds": 2.64, "estimated_voice_end_monotonic": 10963.146521192999 }, "recognition_seconds": 8.202089577000152, "generation_seconds": 7.571548214998984, "first_token_seconds": 2.2086637840002368, "generated_token_count": 32, "eos": false, "speech": { "synthesis_seconds": 6.047646677998273, "playback_seconds": 12.672501942999588, "playback_start_monotonic": 10985.985612601, "playback_end_monotonic": 10998.658114544 }, "interaction": { "cue_to_playback_complete_seconds": 38.151593350999974, "capture_request_to_playback_complete_seconds": 39.840032669999346, "estimated_question_end_to_audio_start_seconds": 22.839091408000968, "estimated_question_end_to_playback_complete_seconds": 35.511593351000556, "note": "ALSA submission/completion wall times; voice endpoint estimated from amplitude, not acoustic loopback", "estimated_question_start_to_playback_complete_seconds": 37.211593350999976 }, "source": "runs/voice_20261023_021906_16176_0" }, { "case": "round1_osaka", "expected_question": "大阪で食べられるおいしいものを教えてください", "recognized": "大阪で食べられるおいしいものを進んでください", "exact_normalized_match": false, "character_error_rate": 0.13636363636363635, "mic_gain_percent": 35, "rms": 257, "peak": 3123, "clipped_fraction": 0.0, "recorded_audio_seconds": 10.08, "capture": { "interaction_start_monotonic": 10998.666878176, "cue_monotonic": 11000.339435857, "capture_finished_monotonic": 11010.632028644, "capture_seconds": 10.292592787000103, "first_voice_offset_seconds": 6.18, "last_voice_offset_seconds": 9.28, "estimated_voice_end_monotonic": 11009.619435857001 }, "recognition_seconds": 10.290834660998371, "generation_seconds": 8.11661832700156, "first_token_seconds": 2.748727604001033, "generated_token_count": 32, "eos": false, "speech": { "synthesis_seconds": 6.500037231000533, "playback_seconds": 13.674811741999292, "playback_start_monotonic": 11035.579807554, "playback_end_monotonic": 11049.254619296 }, "interaction": { "cue_to_playback_complete_seconds": 48.91518343899952, "capture_request_to_playback_complete_seconds": 50.58774111999992, "estimated_question_end_to_audio_start_seconds": 25.960371696999573, "estimated_question_end_to_playback_complete_seconds": 39.635183438998865, "note": "ALSA submission/completion wall times; voice endpoint estimated from amplitude, not acoustic loopback", "estimated_question_start_to_playback_complete_seconds": 42.73518343899952 }, "source": "runs/voice_20261023_021946_16176_1" }, { "case": "round2_paris", "expected_question": "パリはどこにありますか", "recognized": "今ありますか？", "exact_normalized_match": false, "character_error_rate": 0.5454545454545454, "mic_gain_percent": 75, "rms": 362, "peak": 3528, "clipped_fraction": 0.0, "recorded_audio_seconds": 4.64, "capture": { "interaction_start_monotonic": 11422.735841733, "cue_monotonic": 11424.433126743, "capture_finished_monotonic": 11429.22503177, "capture_seconds": 4.791905026999302, "first_voice_offset_seconds": 2.0, "last_voice_offset_seconds": 3.84, "estimated_voice_end_monotonic": 11428.273126743 }, "recognition_seconds": 13.278516200000013, "generation_seconds": 7.3496909589994175, "first_token_seconds": 2.018968722999489, "generated_token_count": 32, "eos": false, "speech": { "synthesis_seconds": 9.212604100001045, "playback_seconds": 19.19436703700012, "playback_start_monotonic": 11459.187629656, "playback_end_monotonic": 11478.381996693 }, "interaction": { "cue_to_playback_complete_seconds": 53.94886994999979, "capture_request_to_playback_complete_seconds": 55.646154960000786, "estimated_question_end_to_audio_start_seconds": 30.914502912999524, "estimated_question_end_to_playback_complete_seconds": 50.108869949999644, "note": "ALSA submission/completion wall times; voice endpoint estimated from amplitude, not acoustic loopback", "estimated_question_start_to_playback_complete_seconds": 51.94886994999979 }, "source": "runs/voice_20261023_022650_18518_0" }, { "case": "round2_osaka", "expected_question": "大阪で食べられるおいしいものを教えてください", "recognized": "大阪で食べられるおいしいものを教えてください", "exact_normalized_match": true, "character_error_rate": 0.0, "mic_gain_percent": 75, "rms": 423, "peak": 3303, "clipped_fraction": 0.0, "recorded_audio_seconds": 6.84, "capture": { "interaction_start_monotonic": 11478.390599183, "cue_monotonic": 11480.062824615, "capture_finished_monotonic": 11487.104993697, "capture_seconds": 7.042169081998509, "first_voice_offset_seconds": 2.36, "last_voice_offset_seconds": 6.04, "estimated_voice_end_monotonic": 11486.102824615002 }, "recognition_seconds": 6.341480450000745, "generation_seconds": 7.991706717000852, "first_token_seconds": 2.63828903600006, "generated_token_count": 32, "eos": false, "speech": { "synthesis_seconds": 3.18348862700077, "playback_seconds": 6.551309234000655, "playback_start_monotonic": 11504.661933059, "playback_end_monotonic": 11511.213242293 }, "interaction": { "cue_to_playback_complete_seconds": 31.15041767799994, "capture_request_to_playback_complete_seconds": 32.82264311000108, "estimated_question_end_to_audio_start_seconds": 18.55910844399841, "estimated_question_end_to_playback_complete_seconds": 25.110417677999067, "note": "ALSA submission/completion wall times; voice endpoint estimated from amplitude, not acoustic loopback", "estimated_question_start_to_playback_complete_seconds": 28.79041767799994 }, "source": "runs/voice_20261023_022745_18518_1" } ], "selected_profile": "balanced: min-active40/max-active700/beam8/lattice2", "rejected": [ "narrow search: worsened real Osaka transcription", "automatic level normalization: no word-accuracy improvement on retained real PCM" ], "unresolved": "Paris recognition still fails; answer factual accuracy not established" }