[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"article-doc:docs\u002Frecorded\u002F3.timestamps-speakers":3},"---\ntitle: Timestamps & Speakers\ndescription: Segment timestamps, confidence, and speaker information.\n---\nTimestamp and speaker information in transcription results.\n\n## Segment timestamps\n\nEach segment includes timestamps:\n\n```json\n{\n  \"id\": 0,\n  \"start_time\": 0.0,\n  \"end_time\": 3.2,\n  \"duration\": 3.2,\n  \"text\": \"The weather is nice today\",\n  \"confidence\": 0.95\n}\n```\n\n| Field | Type | Description |\n|---|---|---|\n| `id` | number | Segment index (from 0) |\n| `start_time` | number | Start time in seconds |\n| `end_time` | number | End time in seconds |\n| `duration` | number | Duration in seconds |\n| `text` | string | Transcribed text |\n| `confidence` | number | Confidence score (0–1, 3 decimal places) |\n\n### Confidence\n\n- Range: 0 to 1\n- 3 decimal places\n- Calculated from `1 - no_speech_prob` when upstream provides it\n\n### Global timestamps\n\nAfter slicing, timestamps are aggregated to global time. `start_time` and `end_time` are relative to the original audio, not chunk-internal time.\n\n## Summary statistics\n\n```json\n{\n  \"summary\": {\n    \"total_duration\": 120.5,\n    \"total_speech_duration\": 95.3,\n    \"overall_speech_ratio\": 0.79,\n    \"num_segments\": 45\n  }\n}\n```\n\n| Field | Description |\n|---|---|\n| `total_duration` | Total audio duration (seconds) |\n| `total_speech_duration` | Actual speech duration (seconds) |\n| `overall_speech_ratio` | Speech ratio |\n| `num_segments` | Number of segments |\n\n## Metadata\n\n```json\n{\n  \"metadata\": {\n    \"language\": \"zh\",\n    \"model\": \"whisper-large-v3-turbo\",\n    \"chunk_count\": 3,\n    \"audio_duration_seconds\": 120.5\n  }\n}\n```\n\n## Speaker diarization\n\nSpeaker diarization is **not available** in the current version and is on the roadmap.\n\n::callout{icon=\"i-lucide-triangle-alert\" color=\"amber\"}\nCurrent segments do not include a `speaker` field. Do not rely on speaker information.\n::\n\n## Precision notes\n\n- Timestamp precision depends on the upstream model and slicing granularity\n- Timestamps at slice boundaries are corrected during aggregation\n- Short speech segments (\u003C1s) may have less precise timestamps\n\n## Related\n\n- [Transcribe Audio](\u002Fdocs\u002Frecorded\u002Ftranscribe-audio) — submit a job\n- [Subtitles](\u002Fdocs\u002Frecorded\u002Fsubtitles) — generate subtitles from timestamps\n- [Batch Jobs API](\u002Fdocs\u002Fapi-reference\u002Fbatch-jobs) — API reference\n",1790059118960]