Transcription

Transcription converts video audio to text. It auto-detects language and returns timed segments, with speaker labels when distinct speakers are detected.

Transcription-only jobs get a higher duration ceiling than jobs that analyze frames — 10 minutes on Free, 30 on Starter, 90 on Pro, 120 on Scale. The higher ceiling applies when the job requests only transcription and/or metadata with no custom queries. Adding summary, brands, or queries drops the job to the standard video ceiling (1 / 3 / 10 / 20 minutes).

curl -X POST https://api.netraflow.com/v1/jobs \
  -H "Content-Type: application/json" \
  -H "X-Api-Key: sk_live_your_key_here" \
  -d '{
    "url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
    "capabilities": ["transcription"]
  }'

{
  "data": {
    "job_id": "job_abc123",
    "status": "completed",
    "results": {
      "transcription": {
        "text": "We're no strangers to love. You know the rules and so do I...",
        "segments": [
          {
            "start": 0.0,
            "end": 4.8,
            "text": "We're no strangers to love",
            "speaker": 0
          },
          {
            "start": 4.8,
            "end": 8.2,
            "text": "You know the rules and so do I",
            "speaker": 0
          }
        ],
        "language": "en",
        "duration_seconds": 212.0,
        "word_count": 423,
        "speakers_detected": 1
      }
    }
  }
}

Response fields

Prop	Type	Description
text	string	Full transcription text.
segments	Segment[]	Timed text segments with optional speaker labels.
word_count	integer	Total number of words transcribed.
words	Word[] (optional)	Word-level timings and confidence. See below.
language	string (optional)	Detected language code (e.g. "en", "es").
duration_seconds	number (optional)	Audio duration in seconds.
speakers_detected	integer (optional)	Number of distinct speakers identified.
no_speech	boolean (optional)	True when the audio contains no detectable speech.

Only text, segments, and word_count are guaranteed. Everything else is optional — see the caveats below.

Videos with no speech

A silent or music-only video is a successful job, not a failure. It returns no_speech: true with text: "" and an empty segments array. On a job that also requests brands, this means no brands can be found in speech; visual detection is unaffected.

Word fields

Prop	Type	Description
word	string	The word as recognized.
start	number	Word start time in seconds.
end	number	Word end time in seconds.
confidence	number (0-1, optional)	Recognition confidence for this word.
punctuated_word	string (optional)	The word with punctuation and casing applied. Fall back to word when absent.

Segment fields

Prop	Type	Description
start	number	Segment start time in seconds.
end	number	Segment end time in seconds.
text	string	Transcribed text for this segment.
speaker	integer	Speaker index (0-based). Present when multiple speakers are detected.

Creator-provided captions

When a YouTube video already ships accurate captions, we use them instead of running speech-to-text — it is faster and costs you less. Caption-derived transcripts are more limited: they return text, segments, language, duration_seconds, and word_count, but no words array, no speakers_detected, and no per-segment speaker.

If you need word-level timings or speaker labels on every job, don't rely on their presence — check for the field. Jobs that also request brands always run full speech-to-text, so they always include words.