Diarization
The Diarization method automatically detects and separates speakers within a single audio channel. This method can handle more complex audio inputs where multiple speakers are mixed in the same channel. However, it is significantly slower, taking 2-3 times longer to process compared to transcription without diarization.
Key Features
Use Case: Suitable for mixed audio inputs where speakers are not on separate channels.
Accuracy: Variable, dependent on the complexity of the audio and number of speakers.
Performance: Slow, taking 2-3 times longer than non-diarized processing.
Recommendation: Use only when Multichannel separation is not feasible.
Transcribe with Diarization
Introduction
This section describes the various ways to use diarization with the Gowajee speech-to-text (STT) API. Diarization is the process of identifying and separating speakers within an audio input. Below are the options for configuring diarization to best suit your needs.
Methods to Enable Diarization
Enable Automatic Diarization
Set diarization to true to enable automatic detection of the number of speakers and speaker separation.
Request:
POST /v1/speech-to-text/${MODEL}/transcribe HTTP/1.1
Host: api.gowajee.ai
Content-Type: application/json
X-Api-Key: ${YOUR_API_KEY}
{
"audioData": "base64_encoded_raw_audio_data",
"diarization": true
}
Response:
{
"type": "ASR_PULSE_WITH_DIARIZE",
"amount": 6.248,
"output": {
"results": [
{
"transcript": "สวัสดีค่ะก่อนอื่นทางเราขอให้คุณยืนยันตัวตนผ่านระบบเสียง",
"startTime": 0,
"endTime": 5.117,
"speaker": "SPEAKER_00"
},
{
"transcript": "ได้เลยครับ",
"startTime": 5.231,
"endTime": 6.248,
"speaker": "SPEAKER_01"
}
],
"duration": 6.248,
"version": "2.2.0"
}
}Specify Number of Speakers
Set diarization to true and define numSpeakers (Integer) to specify the number of speakers in the audio.
Request:
POST /v1/speech-to-text/${MODEL}/transcribe HTTP/1.1
Host: api.gowajee.ai
Content-Type: application/json
X-Api-Key: ${YOUR_API_KEY}
{
"audioData": "base64_encoded_raw_audio_data",
"diarization": true,
"numSpeakers": 2
}
Response:
{
"type": "ASR_PULSE_WITH_DIARIZE",
"amount": 6.248,
"output": {
"results": [
{
"transcript": "สวัสดีค่ะก่อนอื่นทางเราขอให้คุณยืนยันตัวตนผ่านระบบเสียง",
"startTime": 0,
"endTime": 5.117,
"speaker": "SPEAKER_00"
},
{
"transcript": "ได้เลยครับ",
"startTime": 5.231,
"endTime": 6.248,
"speaker": "SPEAKER_01"
}
],
"duration": 6.248,
"version": "2.2.0"
}
}Define Range of Speakers
Set diarization to true and define minSpeakers (Integer) and maxSpeakers (Integer) to automatically detect and separate the number of speakers within the specified range.
Request:
POST /v1/speech-to-text/${MODEL}/transcribe HTTP/1.1
Host: api.gowajee.ai
Content-Type: application/json
X-Api-Key: ${YOUR_API_KEY}
{
"audioData": "base64_encoded_raw_audio_data",
"diarization": true,
"minSpeakers": 1,
"maxSpeakers": 2,
}
Response:
{
"type": "ASR_PULSE_WITH_DIARIZE",
"amount": 6.248,
"output": {
"results": [
{
"transcript": "สวัสดีค่ะก่อนอื่นทางเราขอให้คุณยืนยันตัวตนผ่านระบบเสียง",
"startTime": 0,
"endTime": 5.117,
"speaker": "SPEAKER_00"
},
{
"transcript": "ได้เลยครับ",
"startTime": 5.231,
"endTime": 6.248,
"speaker": "SPEAKER_01"
}
],
"duration": 6.248,
"version": "2.2.0"
}
}Reference Speakers (Recommended)
Define refSpeakers with reference to speaker voices and names. This method improves accuracy by using known speaker samples.
Request:
application/json
POST /v1/speech-to-text/${MODEL}/transcribe HTTP/1.1
Host: api.gowajee.ai
Content-Type: application/json
X-Api-Key: ${YOUR_API_KEY}
{
"audioData": "base64_encoded_raw_audio_data",
"diarization": true,
"refSpeakers": [
{
"name": "Adam",
"audioData": "base64_encoded_voice_of_adam"
},
{
"name": "Bill",
"audioData": "base64_encoded_voice_of_bill"
}
]
}
multipart-form/data
For file uploads using multipart/form-data, the API will use the filename as the speaker name.
POST /v1/speech-to-text/${MODEL}/transcribe HTTP/1.1
Host: api.gowajee.ai
x-api-key: ${YOUR_API_KEY}
Content-Length: ${AUTO}
Content-Type: multipart/form-data; boundary=----WebKitFormBoundary7MA4YWxkTrZu0gW
------WebKitFormBoundary7MA4YWxkTrZu0gW
Content-Disposition: form-data; name="audioData"; filename="audio.wav"
Content-Type: audio/wav
(data)
------WebKitFormBoundary7MA4YWxkTrZu0gW
Content-Disposition: form-data; name="refSpeakers"; filename="Adam.wav"
Content-Type: audio/wav
(data)
------WebKitFormBoundary7MA4YWxkTrZu0gW
Content-Disposition: form-data; name="refSpeakers"; filename="Bill.wav"
Content-Type: audio/wav
(data)
------WebKitFormBoundary7MA4YWxkTrZu0gW
Response:
{
"type": "ASR_PULSE_WITH_DIARIZE",
"amount": 6.248,
"output": {
"results": [
{
"transcript": "สวัสดีครับก่อนอื่นทางเราขอให้คุณยืนยันตัวตนผ่านระบบเสียง",
"startTime": 0,
"endTime": 5.117,
"speaker": "Adam"
},
{
"transcript": "ได้เลยครับ",
"startTime": 5.231,
"endTime": 6.248,
"speaker": "Bill"
}
],
"duration": 6.248,
"version": "2.2.0"
}
}Last updated