tags:
- pyannote
- pyannote-audio
- pyannote-audio-pipeline
- audio
- voice
- speech
- speaker
- speaker-diarization
- speaker-change-detection
- voice-activity-detection
- overlapped-speech-detection
- automatic-speech-recognition
datasets:
- ami
- dihard
- voxconverse
- aishell
- repere
- voxceleb
license: mit
extra_gated_prompt: >-
The collected information will help acquire a better knowledge of
pyannote.audio userbase and help its maintainers apply for grants to improve
it further. If you are an academic researcher, please cite the relevant papers
in your own publications using the model. If you work for a company, please
consider contributing back to pyannote.audio development (e.g. through
unrestricted gifts). We also provide scientific consulting services around
speaker diarization and machine listening.
extra_gated_fields:
Company/university: text
Website: text
I plan to use this model for (task, type of audio data, etc): text
🎹 Speaker diarization
Relies on pyannote.audio 2.1.1: see installation instructions.
TL;DR
# 1. visit hf.co/pyannote/speaker-diarization and accept user conditions
# 2. visit hf.co/pyannote/segmentation and accept user conditions
# 3. visit hf.co/settings/tokens to create an access token
# 4. instantiate pretrained speaker diarization pipeline
from pyannote.audio import Pipeline
pipeline = Pipeline.from_pretrained("pyannote/[email protected]",
use_auth_token="ACCESS_TOKEN_GOES_HERE")
# apply the pipeline to an audio file
diarization = pipeline("audio.wav")
# dump the diarization output to disk using RTTM format
with open("audio.rttm", "w") as rttm:
diarization.write_rttm(rttm)
Advanced usage
In case the number of speakers is known in advance, one can use the num_speakers option:
diarization = pipeline("audio.wav", num_speakers=2)
One can also provide lower and/or upper bounds on the number of speakers using min_speakers and max_speakers options:
diarization = pipeline("audio.wav", min_speakers=2, max_speakers=5)
Benchmark
Real-time factor
Real-time factor is around 2.5% using one Nvidia Tesla V100 SXM2 GPU (for the neural inference part) and one Intel Cascade Lake 6248 CPU (for the clustering part).
In other words, it takes approximately 1.5 minutes to process a one hour conversation.
Accuracy
This pipeline is benchmarked on a growing collection of datasets.
Processing is fully automatic:
- no manual voice activity detection (as is sometimes the case in the literature)
- no manual number of speakers (though it is possible to provide it to the pipeline)
- no fine-tuning of the internal models nor tuning of the pipeline hyper-parameters to each dataset
... with the least forgiving diarization error rate (DER) setup (named "Full" in this paper):
- no forgiveness collar
- evaluation of overlapped speech
| Benchmark | DER% | FA% | Miss% | Conf% | Expected output | File-level evaluation |
|---|---|---|---|---|---|---|
| AISHELL-4 | 14.09 | 5.17 | 3.27 | 5.65 | RTTM | eval |
| Albayzin (RTVE 2022) | 25.60 | 5.58 | 6.84 | 13.18 | RTTM | eval |
| AliMeeting (channel 1) | 27.42 | 4.84 | 14.00 | 8.58 | RTTM | eval |
| AMI (headset mix, only_words) | 18.91 | 4.48 | 9.51 | 4.91 | RTTM | eval |
| AMI (array1, channel 1, only_words) | 27.12 | 4.11 | 17.78 | 5.23 | RTTM | eval |
| CALLHOME (part2) | 32.37 | 6.30 | 13.72 | 12.35 | RTTM | eval |
| DIHARD 3 (Full) | 26.94 | 10.50 | 8.41 | 8.03 | RTTM | eval |
| Ego4D v1 (validation) | 63.99 | 3.91 | 44.42 | 15.67 | RTTM | eval |
| REPERE (phase 2) | 8.17 | 2.23 | 2.49 | 3.45 | RTTM | eval |
| This American Life | 20.82 | 2.03 | 11.89 | 6.90 | RTTM | eval |
| VoxConverse (v0.3) | 11.24 | 4.42 | 2.88 | 3.94 | RTTM | eval |
Technical report
This report describes the main principles behind version 2.1 of pyannote.audio speaker diarization pipeline.
It also provides recipes explaining how to adapt the pipeline to your own set of annotated data. In particular, those are applied to the above benchmark and consistently leads to significant performance improvement over the above out-of-the-box performance.
Support
For commercial enquiries and scientific consulting, please contact me.
For technical questions and bug reports, please check pyannote.audio Github repository.
Citations
@inproceedings{Bredin2021,
Title = {{End-to-end speaker segmentation for overlap-aware resegmentation}},
Author = {{Bredin}, Herv{\'e} and {Laurent}, Antoine},
Booktitle = {Proc. Interspeech 2021},
Address = {Brno, Czech Republic},
Month = {August},
Year = {2021},
}
@inproceedings{Bredin2020,
Title = {{pyannote.audio: neural building blocks for speaker diarization}},
Author = {{Bredin}, Herv{\'e} and {Yin}, Ruiqing and {Coria}, Juan Manuel and {Gelly}, Gregory and {Korshunov}, Pavel and {Lavechin}, Marvin and {Fustes}, Diego and {Titeux}, Hadrien and {Bouaziz}, Wassim and {Gill}, Marie-Philippe},
Booktitle = {ICASSP 2020, IEEE International Conference on Acoustics, Speech, and Signal Processing},
Address = {Barcelona, Spain},
Month = {May},
Year = {2020},
}