AldeaSpeaker diarization
Identify and label individual speakers with diarize=true
Speaker diarization is the process of identifying "who spoke when". It breaks down an audio stream into time-labelled segments, attributing each to a distinct speaker. Diarization answers which part of the audio belongs to which speaker without necessarily knowing their real identities. It's helpful when working with meeting transcripts, call center recordings, and podcast transcription to produce labeled output like:
Speaker 1 [0.00 - 0.08]: "Welcome to our weekly meeting..."
Speaker 2 [0.09 - 0.36]: "Thank you, I'll go first with my updates..."Usage
To enable diarization in your transcript, add diarize=true as a query parameter to a pre-recorded transcription request:
curl -X POST "https://api.aldea.ai/v1/listen?diarize=true" \
-H "Authorization: Bearer YOUR_ALDEA_API_KEY" \
-H "timestamps: true" \
--data-binary @interview.wavYou can combine it with other parameters:
curl -X POST "https://api.aldea.ai/v1/listen?language=es&diarize=true&smart_format=true" \
-H "Authorization: Bearer YOUR_ALDEA_API_KEY" \
-H "timestamps: true" \
--data-binary @interview.mp3Response
When diarization is enabled, each word object in the response includes a speaker field with an integer identifier:
{
"results": {
"channels": [{
"alternatives": [{
"transcript": "How are you doing today? I'm doing great, thanks.",
"confidence": 0.88,
"words": [
{ "word": "How", "start": 0.08, "end": 0.24, "speaker": 0 },
{ "word": "are", "start": 0.24, "end": 0.36, "speaker": 0 },
{ "word": "you", "start": 0.36, "end": 0.48, "speaker": 0 },
{ "word": "doing", "start": 0.48, "end": 0.72, "speaker": 0 },
{ "word": "today?", "start": 0.72, "end": 1.04, "speaker": 0 },
{ "word": "I'm", "start": 1.28, "end": 1.44, "speaker": 1 },
{ "word": "doing", "start": 1.44, "end": 1.68, "speaker": 1 },
{ "word": "great,", "start": 1.68, "end": 1.92, "speaker": 1 },
{ "word": "thanks.", "start": 1.92, "end": 2.16, "speaker": 1 }
]
}]
}]
}
}Speaker indices start at 0 and are assigned in the order speakers are detected. Word timestamps are required for diarization to work.