gryannote: a speaker diarization labeling tool

gryannote is a collection of Gradio custom components focusing on the labeling of speaker diarization data. Integrated with the pyannote speaker diarization ecosystem, it allows to build web applications to load pretrained pyannote pipelines and customize their hyper-parameters, upload or record an audio file, process it with the pipeline, visualize and interact with its outputs, correct them if needed, and export the final annotation in RTTM format. Each of these components can be used independently from each other.

Available Gradio custom components

Here is the list of Gradio custom components integrated in gryannote

• gryannote_audio
• gryannote_pipeline
• gryannote_rttm

Installation

pip install gryannote

Usage

The following code snippet show how to use the gryannote_audio component with a pyannote pipeline in just a few lines of code. You can find a complete example that uses the three component in app.py script.

import gradio as gr
from gryannote_audio import AudioLabeling
from pyannote.audio import Pipeline

audio_labeling = AudioLabeling(type="filepath", interactive=True)


def apply_pipeline(audio):
    pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization-3.1")
    annotations = pipeline(audio)
    return (audio, annotations)

demo = gr.Interface(apply_pipeline, inputs=audio_labeling, outputs=audio_labeling)

demo.launch()

Interface

Launching demo/app.py script will generate the following interface. This interface uses the three gryannote components. More details about these components and their interface can be found in their respective README.

RTTM annotations in RTTM component are dynamically updated according to the audio labeling made in the audio component.

Try it!

A gryannote app can be run in this Hugging Face space

Build components from source

To build the gryannote components from source (e.g., for development purposes), follow these steps:

First, make sure your environment is ready for Gradio custom component development.

Clone the gryannote repository:

git clone git@github.com:clement-pages/gryannote.git

Navigate to a component directory (for example the audio component) and install it using the Gradio CLI:

cd gryannote/audio
gradio cc install

To start a development server and test the component in real time:

gradio cc dev

Once you're done with development, build the component:

gradio cc build

For more information on gradio cc commands and custom component development, refer to the official Gradio documentation.

Citation

If you are using gryannote, please use the following citation:

@inproceedings{pages24_interspeech,
  title     = {Gryannote open-source speaker diarization labeling tool},
  author    = {Clément Pages and Hervé Bredin},
  year      = {2024},
  booktitle = {Interspeech 2024},
  pages     = {3650--3651},
}