Output Format Reference

The extract command produces two output files.

CSV (output.csv)

Contains interleaved segment-level and word-level rows. See example.

Column	Description
type	"segment" or "word"
speaker	Speaker ID (e.g. SPEAKER_00)
start	Start time in seconds
end	End time in seconds
text	Full segment text
word	Individual word (word rows only)
word_position	Word index within segment (word rows only)
confidence	Confidence score (0–1)
overlap_duration	Speaker overlap duration in seconds

Segment rows and their corresponding word rows are interleaved — each segment row is immediately followed by its word rows.

Metadata (output_metadata.json)

See example.

{
  "audio_file": "interview.mp3",
  "language": "en",
  "language_probability": 0.99,
  "duration": 527.19,
  "total_segments": 87,
  "total_words": 1234,
  "speakers": ["SPEAKER_00", "SPEAKER_01"],
  "processing_timestamp": "2026-03-05 22:00:00"
}

Field	Description
audio_file	Path to the input audio file
language	Detected language code (e.g. "en")
language_probability	Confidence of language detection (0–1)
duration	Total audio duration in seconds
total_segments	Number of speaker segments
total_words	Total word count across all segments
speakers	List of detected speaker IDs
processing_timestamp	When the file was processed