Repolode
Submit
pyannote

pyannote/pyannote-audio

Open-source Python toolkit for speaker diarization, built on PyTorch, with pretrained pipelines for detecting who spoke when in audio.

View on GitHubWebsite
Stars
10,671
Forks
1,123
Open issues
48
Last commit
14 hours ago

What is it?

pyannote.audio is a Python toolkit for speaker diarization, the task of segmenting an audio recording by speaker so you can tell who spoke when. It is built on PyTorch and ships pretrained pipelines and models that can be run locally or fine-tuned on your own data.

The toolkit covers the building blocks of diarization, including speech activity detection, speaker change detection, overlapped speech detection, and speaker embedding. The current open-source pipeline, community-1, is loaded from Hugging Face with a short Python call and returns timed speaker turns. A premium pipeline, precision-2, runs on pyannoteAI's servers through an API key, and an optional telemetry setting can be turned off with an environment variable.

Training supports multi-GPU runs through pytorch-lightning. The project also includes tutorials on applying, adapting and training pipelines and models.

Key features

  • 01Pretrained speaker diarization pipelines loadable from Hugging Face
  • 02Speech activity, speaker change, and overlapped speech detection models
  • 03Speaker embedding extraction for verification and recognition
  • 04Fine-tuning on your own data with multi-GPU training via pytorch-lightning
  • 05Optional hosted premium pipeline through the pyannoteAI API

Use cases

  • Transcribing meetings or interviews with speaker labels
  • Analyzing call-center or podcast recordings by speaker
  • Researching or benchmarking speaker diarization models
  • Adapting a diarization model to a specific language or domain

Quick start

  1. 1

    Make sure ffmpeg is installed on your machine

  2. 2

    Install the library with uv (recommended)

    uv add pyannote.audio
  3. 3

    Accept the user conditions for pyannote/speaker-diarization-community-1 on Hugging Face and create an access token

  4. 4

    Load the community-1 pipeline and run it on an audio file

Things to know

  • The open-source community-1 pipeline requires accepting model conditions and a Hugging Face access token.
  • Running the pipeline is faster on a GPU, and the README benchmarks use an NVIDIA H100.
  • The premium precision-2 service requires a pyannoteAI API key and runs on pyannoteAI servers for hosted use.

Topics

#speaker-diarization#speech-processing#pytorch#voice-activity-detection#speaker-recognition#pretrained-models#overlapped-speech-detection#speaker-change-detection#speaker-embedding#speaker-verification

FAQ

What is pyannote-audio?
Open-source Python toolkit for speaker diarization, built on PyTorch, with pretrained pipelines for detecting who spoke when in audio. pyannote.audio is a Python toolkit for speaker diarization, the task of segmenting an audio recording by speaker so you can tell who spoke when. It is built on PyTorch and ships pretrained pipelines and models that can be run locally or fine-tuned on your own data.
Is pyannote-audio free to use? What license does it have?
pyannote-audio is open source under the MIT license. Check the license terms before using it in commercial products.
Is pyannote-audio still maintained?
Actively maintained. The last commit to pyannote-audio was on Oct 20, 2026.
What can I use pyannote-audio for?
Transcribing meetings or interviews with speaker labels Analyzing call-center or podcast recordings by speaker Researching or benchmarking speaker diarization models Adapting a diarization model to a specific language or domain
How do I install pyannote-audio?
Make sure ffmpeg is installed on your machine Install the library with uv (recommended): uv add pyannote.audio Accept the user conditions for pyannote/speaker-diarization-community-1 on Hugging Face and create an access token Load the community-1 pipeline and run it on an audio file
Where can I get pyannote-audio?
The source code and installation instructions are on GitHub: https://github.com/pyannote/pyannote-audio

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects