POST
/
v1
/
diarize
Diarize audio
curl --request POST \
  --url https://api.pyannote.ai/v1/diarize \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '{
  "url": "https://example.com/audio.wav",
  "webhook": "https://example.com/webhook",
  "model": "precision-2",
  "numSpeakers": 2,
  "minSpeakers": 1,
  "maxSpeakers": 4,
  "turnLevelConfidence": true,
  "exclusive": true,
  "confidence": true
}'
{
  "jobId": "3c8a89a5-dcc6-4edb-a75d-ffd64739674d",
  "status": "created"
}
This endpoint allows you to create a new diarization from a remote audio URL.

​
Diarize a remote file (from a URL)

If you have a media file accessible via a URL, you can provide the URL to the file in the request body. Typically you would use this method with a file stored in a cloud storage service such as Amazon S3. If bucket or object is private, you can follow this tutorial and create a signed url to give temporary access. For performance reasons, the input audio must respect the following limits:
  • Maximum duration of 24hs
  • Maximum file size of 1GiB
Make sure the URL to the file is publicly and directly accessible, otherwise our endpoint won’t be able to read the file.
Some cloud storage services provide an indirect URL with a confirmation page to access the file. Make sure you provide a direct link to the file instead.

​
Diarization of a Local File

If you have a media file stored locally, you can upload it to our temporary media storage and provide the URL to the file in the request body. Typically you would use this method if you have a file stored on a local machine or device. Follow the How to upload files tutorial to upload your file.

​
Choosing models

As new models are released, you can specify which model to use by adding the model parameter to your request body. The model parameter allows you to select a specific model, which can be useful for optimizing performance or accuracy based on your specific use case. The available models are:
  • precision-1: The current precision model (default)
  • precision-2: Our latest precision model (37% better speaker assignment)
We highly recommend switching your new jobs to precision-2, as it will become the new default in the near future.
Voiceprints generated with precision-1 and precision-2 are fully compatible. You can use voiceprints from either model interchangeably for identification tasks.

​
Number of speakers

By default, pyannoteAI will automatically detect the number of speakers present in the input audio, and there is no limit as to how many speakers can be detected. It is common to diarize audio where the number of speakers is already known to be 2 (e.g. phone conversations or interviews). If this is your use case, you can add "numSpeakers": 2 to the body of your request. This will detect exactly 2 speakers, which typically results in better overall diarization performance. If the number of speakers is unknown but falls within a certain range, or if you want to set a limit, you can use the “minSpeakers” and “maxSpeakers” parameters. For example, specifying “minSpeakers”: 2, “maxSpeakers”: 4 will restrict detection to between two and four speakers.
Setting numSpeakers=2 is equivalent to minSpeakers=2 and maxSpeakers=2.numSpeakers can’t be used together with minSpeakers or maxSpeakers.If both minSpeakers and maxSpeakers are set, minSpeakers must be less than or equal to maxSpeakers.

​
Diarization without overlapping speech

If you need results where speech from different speakers does not overlap, you can enable exclusive diarization by adding "exclusive": true to your request body. This will include exclusive diarization values in the output (equivalent to diarization but without overlapping speech).

​
Confidence Scores

Confidence scores provide a measure of the certainty of the diarization model in its predictions. These scores range from 0 to 100, with higher values indicating greater confidence. To include confidence scores in your diarization results, add "confidence": true to your request body. When enabled, the job output will include:
  • The confidence object containing:
    • score, an array with the confidence score for each sample
    • resolution, indicating the time interval in seconds between confidence score samples
For example, if the resolution is 0.02, it means each confidence score represents a 20-millisecond interval in the audio. Confidence scores can be particularly useful for:
  • Identifying segments where the model is less certain, which may benefit from manual review
  • Filtering results based on a confidence threshold for higher accuracy
  • Analyzing the overall reliability of the diarization for a given audio file
By default, confidence scores are not included in the output. To enable them, you must explicitly set "confidence": true in your request.

​
Turn-level confidence scores

To include turn-level confidence scores in your results, add "turnLevelConfidence": true to your request body. When enabled, the job output will include a "confidence" object for each diarization segment. The object contains each speaker as the key and a number from 0-100 indicating the confidence score for each speaker assignment.

​
Getting results using a webhook

If the optional webhook param is set, the results will be sent to the provided URL. The webhook URL is where any update related to the diarization job will be sent. The job output will be sent as a JSON object in the request body.
Make sure the webhook URL is publicly accessible, with a valid SSL certificate, otherwise our server cannot send the output
Please visit the Receiving webhooks documentation for more information.

​
Polling for job results

As an alternative to webhooks, you can poll the get job endpoint to retrieve job results. To poll for job results, use the status field in the job response to check if the job is completed. Once the job is completed, you can retrieve the job results by accessing the output field.
The output field contains the job results and is only available when the job status is succeeded. Job results are automatically deleted after 24 hours of job completion.
A job is completed when the status field is either succeeded, failed, or canceled.
Be cautious of rate limits when polling. Excessive requests to check job status can lead to rate limiting, potentially disrupting your workflow. Check the rate limits documentation for more information.

Authorizations

​
Authorization
string
header
required

Bearer authentication header of the form Bearer <token>, where <token> is your auth token.

Body

application/json
​
url
string
required

URL of the audio file to be processed

Example:

"https://example.com/audio.wav"

​
webhook
string

Webhook URL to receive results when job is completed (optional)

Example:

"https://example.com/webhook"

​
model
enum<string> | null
default:precision-1
Available options:
precision-1,
precision-2
Example:

"precision-2"

​
numSpeakers
number

Number of speakers. Only use if the number of speakers is known in advance. Number of speakers is detected automatically if not provided. Setting this value results in better overall diarization performance. In rare cases where we cannot honor this request (e.g. short files and large number of speakers), a warning will be added to the output. Equivalent to sending minSpeakers==maxSpeakers

Required range: x >= 1
Example:

2

​
minSpeakers
number

Minimum number of speakers (must be <= maxSpeakers if both are set)

Required range: x >= 1
Example:

1

​
maxSpeakers
number

Maximum number of speakers (must be >= minSpeakers if both are set)

Required range: x >= 1
Example:

4

​
turnLevelConfidence
boolean | null
default:false

Includes turn-level confidence values in the output.

Example:

true

​
exclusive
boolean
default:false

Includes exclusive diarization values in the output (equivalent to diarization but without overlapping speech).

Example:

true

​
confidence
boolean
default:false

Include confidence values in the output

Example:

true

Response

​
jobId
string
required

ID of the job

Example:

"3c8a89a5-dcc6-4edb-a75d-ffd64739674d"

​
status
enum<string>
required

Status of the job

Available options:
pending,
created,
succeeded,
canceled,
failed,
running
Example:

"created"