Diffusion Model for Non-Autoregressive Speech Recognition

Environment

pip3 install -r requirements.txt

Training and Inference

Select Language

• English

  cd English
  

• Chinese

  cd Chinese
  

Select Acoustic Encoder

• wav2vec 2.0

  cd wav2vec2
  

• HuBERT

  cd hubert
  

• WavLM

  cd wavlm
  

Download Dataset

• English

  bash dataset.sh
  

• Chinese

  Download dataset from Common Voice by selecting Chinese (Taiwan) and Common Voice Corpus 17.0

  bash dataset.sh
  

Extract Feature

python3 extract.py

Training

python3 train.py

Inference

• Without fast sampling

  python3 inference.py
  

• With fast sampling

  python3 inference_2.py --C 0.99 --T 5