diff --git a/README.md b/README.md index 9f5eae4e..f0a9813d 100644 --- a/README.md +++ b/README.md @@ -1,7 +1,11 @@ -Using `pyannote.audio` open-source toolkit in production? -Consider switching to [ `pyannoteAI`](https://www.pyannote.ai) for better and faster options. +

+ +

+ +
+

pyannote speaker diarization toolkit

+
-# `pyannote` speaker diarization toolkit `pyannote.audio` is an open-source toolkit written in Python for speaker diarization. Based on [PyTorch](https://pytorch.org) machine learning framework, it comes with state-of-the-art [pretrained models and pipelines](https://hf.co/pyannote), that can be further finetuned to your own data for even better performance. @@ -17,7 +21,7 @@ Consider switching to [ `pyannote.audio` open-source speaker diarization pipeline +## `community-1` open-source speaker diarization 1. Make sure [`ffmpeg`](https://ffmpeg.org/) is installed on your machine (needed by [`torchcodec`](https://docs.pytorch.org/torchcodec/) audio decoding library) 2. Install with [`uv`](https://docs.astral.sh/uv/)`add pyannote.audio` (recommended) or `pip install pyannote.audio` @@ -29,7 +33,7 @@ import torch from pyannote.audio import Pipeline from pyannote.audio.pipelines.utils.hook import ProgressHook -# Open-source pyannote speaker diarization pipeline +# Community-1 open-source speaker diarization pipeline pipeline = Pipeline.from_pretrained( "pyannote/speaker-diarization-community-1", token="HUGGINGFACE_ACCESS_TOKEN") @@ -51,15 +55,15 @@ for turn, _, speaker in output.speaker_diarization.itertracks(yield_label=True): ``` -## `pyannoteAI` premium speaker diarization pipeline +## `precision-2` premium speaker diarization 1. Create pyannoteAI API key at [`dashboard.pyannote.ai`](https://dashboard.pyannote.ai) -2. Enjoy 150 hours of free credits! +2. Enjoy free credits! ```python from pyannote.audio import Pipeline -# Premium pyannoteAI speaker diarization service +# Precision-2 premium speaker diarization service pipeline = Pipeline.from_pretrained( "pyannote/speaker-diarization-precision-2", token="PYANNOTEAI_API_KEY") @@ -77,34 +81,33 @@ for turn, _, speaker in diarization.itertracks(yield_label=True): Visit [`docs.pyannote.ai`](https://docs.pyannote.ai) to learn about other pyannoteAI features (voiceprinting, confidence scores, ...) ## Benchmark -Out of the box, `pyannote.audio` speaker diarization [pipeline v4.0](https://hf.co/pyannote/speaker-diarization-community-1) is expected to be much better than v3.1. - `pyannoteAI` premium models are even better (and also 2x faster). `labs` model is currently in private beta. - - -| Benchmark (last updated in 2025-08) |
v3.1
|
v4.0
|
API
|
labs
| -| --------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------ | -------------------------------------------------| ------------------------------------------------ | --- | -| [AISHELL-4](https://arxiv.org/abs/2104.03603) | 12.2 | 11.7 | 11.8 | 11.4 | -| [AliMeeting](https://www.openslr.org/119/) (channel 1) | 24.5 | 20.3 | 16.3 | 15.2 | -| [AMI](https://groups.inf.ed.ac.uk/ami/corpus/) (IHM) | 18.8 | 17.0 | 13.2 | 12.9 | -| [AMI](https://groups.inf.ed.ac.uk/ami/corpus/) (SDM) | 22.7 | 19.9 | 15.8 | 15.6 | -| [AVA-AVD](https://arxiv.org/abs/2111.14448) | 49.7 | 44.6 | 40.7 | 37.1 | -| [CALLHOME](https://catalog.ldc.upenn.edu/LDC2001S97) ([part 2](https://github.com/BUTSpeechFIT/CALLHOME_sublists/issues/1)) | 28.5 | 26.7 | 17.6 | 16.6 | -| [DIHARD 3](https://catalog.ldc.upenn.edu/LDC2022S14) ([full](https://arxiv.org/abs/2012.01477)) | 21.4 | 20.2 | 15.7 | 14.7 | -| [Ego4D](https://arxiv.org/abs/2110.07058) (dev.) | 51.2 | 46.8 | 44.7 | 39.0 | -| [MSDWild](https://github.com/X-LANCE/MSDWILD) | 25.4 | 22.8 | 17.9 | 17.3 | -| [RAMC](https://www.openslr.org/123/) | 22.2 | 20.8 | 10.6 | 10.5 | -| [REPERE](https://www.islrn.org/resources/360-758-359-485-0/) (phase2) | 7.9 | 8.9 | 7.3 | 7.4 | -| [VoxConverse](https://github.com/joonson/voxconverse) (v0.3) | 11.2 | 11.2 | 9.0 | 8.5 | +| Benchmark (last updated in 2025-09) | `legacy` (3.1)| `community-1` | `precision-2` | +| --------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------ | -------------------------------------------------| ------------------------------------------------ | +| [AISHELL-4](https://arxiv.org/abs/2104.03603) | 12.2 | 11.7 | 11.8 | +| [AliMeeting](https://www.openslr.org/119/) (channel 1) | 24.5 | 20.3 | 16.3 | +| [AMI](https://groups.inf.ed.ac.uk/ami/corpus/) (IHM) | 18.8 | 17.0 | 13.2 | +| [AMI](https://groups.inf.ed.ac.uk/ami/corpus/) (SDM) | 22.7 | 19.9 | 15.8 | +| [AVA-AVD](https://arxiv.org/abs/2111.14448) | 49.7 | 44.6 | 40.7 | +| [CALLHOME](https://catalog.ldc.upenn.edu/LDC2001S97) ([part 2](https://github.com/BUTSpeechFIT/CALLHOME_sublists/issues/1)) | 28.5 | 26.7 | 17.6 | +| [DIHARD 3](https://catalog.ldc.upenn.edu/LDC2022S14) ([full](https://arxiv.org/abs/2012.01477)) | 21.4 | 20.2 | 15.7 | +| [Ego4D](https://arxiv.org/abs/2110.07058) (dev.) | 51.2 | 46.8 | 44.7 | +| [MSDWild](https://github.com/X-LANCE/MSDWILD) | 25.4 | 22.8 | 17.9 | +| [RAMC](https://www.openslr.org/123/) | 22.2 | 20.8 | 10.6 | +| [REPERE](https://www.islrn.org/resources/360-758-359-485-0/) (phase2) | 7.9 | 8.9 | 7.3 | +| [VoxConverse](https://github.com/joonson/voxconverse) (v0.3) | 11.2 | 11.2 | 9.0 | __[Diarization error rate](http://pyannote.github.io/pyannote-metrics/reference.html#diarization) (in %, the lower, the better)__ -| Benchmark (last updated in 2025-08) | | | Speed up +Compared to the [`3.1`](https://hf.co/pyannote/speaker-diarization-3.1) legacy pipeline, [`community-1`](https://hf.co/pyannote/speaker-diarization-community-1) brings significant improvement in terms of speaker counting and assignment. +[`precision-2`](https://www.pyannote.ai/blog/precision-2) premium pipeline further improves accuracy as well as processing speed (in its self-hosted version). + +| Benchmark (last updated in 2025-09) | `community-1` | `precision-2` | Speed up | -------------- | ----------- | ----------- | ------ | | [AMI](https://groups.inf.ed.ac.uk/ami/corpus/) (IHM), ~1h files | 31s per hour of audio | 14s per hour of audio | 2.2x faster | [DIHARD 3](https://catalog.ldc.upenn.edu/LDC2022S14) ([full](https://arxiv.org/abs/2012.01477)), ~5min files | 37s per hour of audio | 14s per hour of audio | 2.6x faster -__Processing speed on a NVIDIA H100 80GB HBM3__ +__Self-hosted speed on a NVIDIA H100 80GB HBM3__ ## Telemetry