Learning Adapters for Code-Switching Speech Recognition

This repository is developed under huggingface Framework

Requirements

python 3.8.10

• datasets >= 1.18.0
• torch >= 1.5
• torchaudio
• librosa
• jiwer
• evaluate
• numpy
• pandas
• jieba
• editdistance
• tensorboard
• fairscale
• seaborn
• accelerate
• spacy

Installatation

1. Install huggingface

> cd transformers
> pip install -e .

dataset and pretrained weight

• dataset

MLLAB-public (\\mllab.asuscomm.com)W:\Chun-Yi_He\ASR_data\NTUT\dataset_NTUT
MLLAB-public (\\mllab.asuscomm.com)W:\Chun-Yi_He\ASR_data\ASCEND

• pretrained weight

MLLAB-public (\\mllab.asuscomm.com)W:\Chun-Yi_He\pretrained_weight

FIle structure

|_ /ASCEND/
  |_ dataset_NTUT (NTUT AB01 dataset)  
  |_ waves (ASCEND dataset)
  |_ pretrained_weight

Model training

> cd examples/pytorch/speech-recognition/ASCEND/
> pip install -r requirements.txt
> bash run_train.sh

Model inference

python inference.py