Skip to content

Latest commit

 

History

History
35 lines (23 loc) · 1.76 KB

readme_en.md

File metadata and controls

35 lines (23 loc) · 1.76 KB
Raw

The ASR based on the SPN

Introduction

This project is a SPN solution for the re-micro-cup digital one-game problem - that is, the use of SPN network to build an isolated word speech recognition model. The project file gives the code for waveform data pre-processing and the code for model training.

The project catalog

----ASR\
    |----PreProcess\           
    |    |----preprocess.py    
    |----Train\                
    |    |----datasplit.py     
    |    |----dtwsplit.py      
    |    |----train_spn.ipynb  
    |----requirements.txt       
    |----readme.md

Instructions for use

  1. Environmental requirements: Python3.8, you need to install the pypi package in requirement.txt, you can install it directly through the command pip install -r requirements.txt

  2. Data preparation: Please prepare the data collection of Google Speech Command and place it in folders according to the label;

  3. Data preprocessing: open the file preprocess.py to modify the path of the input waveform data file and configure the output path;

  4. Model training: When performing model training, configure the path in each file in the ./Train/ directory, and then execute dtwsplit.py, datasplit.pyandtrain_spn.ipynbfiles in sequence. Among them, please use jupyter lab/notebook to opentrain_spn.ipynb`, and execute them in order from top to bottom.

  5. Save model: There is a function to export SPN model in train_spn.ipynb, you only need to configure the path.

other instructions

Since the complete training data is relatively large, only a small data set is placed in the warehouse. If necessary, please download Google Speech Command by yourself or download it from Baidu Cloud Disk. (Link: https://pan.baidu.com/s/1fXTGaAYHVPDtipNF287x-w Extraction code: qwi2 )