UniSpeech/UniSpeech-SAT

UniSpeech-SAT

This is the official implementation of paper "UniSpeech-SAT: Universal Speech Representation Learning with Speaker Aware Pre-Training"(ICASSP 2022 Submission). The implementation mainly based on fairseq codebase.

Requirements and Installation

• Pytorch >= 1.6.0
• python version >= 3.7

cd UniSpeech/UniSpeech-SAT
pip install --editable ./ --user

Pre-trained models

Model	Dataset	Model
UniSpeech-SAT Base	960 hrs LibriSpeech	download
UniSpeech-SAT Base+	60k hrs Libri-Light + 10k hrs GigaSpeech + 24k hrs VoxPopuli	download
UniSpeech-SAT Large	60k hrs Libri-Light + 10k hrs GigaSpeech + 24k hrs VoxPopuli	download

Load pretrained model

Example usage:

import torch
import fairseq

cp_path = '/path/to/wav2vec.pt'
model, cfg, task = fairseq.checkpoint_utils.load_model_ensemble_and_task([cp_path])
model = model[0]
model.remove_pretraining_modules()
model.eval()

wav_input_16khz = torch.randn(1,10000)
f = model.feature_extractor(wav_input_16khz)

Results on SUPERB

Citation

If you find our work useful, please cite our paper.