How do I get timestamps and speaker labels from a transcription API?

To get timestamps and speaker labels from a transcription API, you usually need to enable the right options in the request and then parse the structured response.

1) Enable timestamps

Most transcription APIs support one or more of these:

• Word-level timestamps: start/end time for each word
• Segment-level timestamps: time for each chunk/utterance
• Sentence-level timestamps: less common, often derived from segments

Look for request options like:

• timestamps: true
• word_timestamps: true
• return_timestamps: true
• granularity: word

2) Enable speaker diarization

Speaker labels usually come from diarization, which identifies who spoke when. Common options:

• diarization: true
• speaker_labels: true
• enable_speaker_diarization: true
• diarize: true

Some APIs also let you set:

• max_speakers
• min_speakers

3) Make sure your API supports both

Not every transcription API returns both by default. Some provide:

• timestamps only
• speaker labels only
• timestamps + diarization
• or require a separate diarization pipeline

4) Example response shape

A typical response might look like:

{
  "segments": [
    {
      "speaker": "SPEAKER_00",
      "start": 0.0,
      "end": 4.2,
      "text": "Hello everyone."
    },
    {
      "speaker": "SPEAKER_01",
      "start": 4.2,
      "end": 7.8,
      "text": "Hi, thanks for joining."
    }
  ],
  "words": [
    {
      "word": "Hello",
      "start": 0.0,
      "end": 0.5,
      "speaker": "SPEAKER_00"
    }
  ]
}

5) Example using an API request

Generic example

{
  "audio_url": "https://example.com/audio.mp3",
  "timestamps": true,
  "diarization": true,
  "max_speakers": 2
}

6) Parse the output

If the API returns JSON, you can extract:

• start / end for timestamps
• speaker for labels
• text for the transcript

7) Important notes

• Speaker labels are usually estimated, not true identities. You’ll often get labels like SPEAKER_00, SPEAKER_01.
• Diarization works best with:
  • clear audio
  • minimal overlap
  • one speaker at a time
• If speakers talk over each other, labels may be imperfect.

If you want, I can show you how to do this with a specific API like OpenAI, AssemblyAI, Deepgram, Google Speech-to-Text, or AWS Transcribe.