Real-Time Diarization

This project implements dynamic and, to some extend, static quantization (with and without QAT) of the diart framework.

Note that the latency speed-ups which occur due to our quantization are rather negligible.

The files you will be most interested in are quantize_models.py and diart_benchmark.py. The former file is responsible for running the quantization and performing single-inference latency benchmarks on the dynamically quantized models. The latter file implements a script to run benchmarks on all implmented models. This script can be widely configured by the usage of CLI arguments.

Preliminaries

Accept the terms and conditions of the following models

• pyannote/segmentation: https://huggingface.co/pyannote/segmentation
• pyannote/segmentation-3.0: https://huggingface.co/pyannote/segmentation-3.0
• pyannote/embedding: https://huggingface.co/pyannote/embedding

Create a huggingface access token: https://hf.co/settings/tokens. Then use huggingface-cli to log into huggingface.

On some devices you might have to run:

sudo apt-get install libportaudio2
pip install sounddevice

Set-Up

Python Environment

Run one of the two commands

conda env create --file="environment.yml"

pip install -r requirements.txt

Dataset

Download the AMI dataset

git submodule update --init --recursive
cd AMI-diarization-setup/pyannote
sh download_ami.sh

Use the following script to copy the AMI files into a custom directory such that they can be used within diart (diart does not support the default file structure from pyannote)

sh create_diart_ami_dataset.sh