ikegami-yukino / csj-eval

For evaluating speech recognition system using the Corpus of Spontaneous Japanese (CSJ)

csj-eval

This repository aims at evaluating speech recognition system using the Corpus of Spontaneous Japanese (CSJ)

CSJ is available at https://clrd.ninjal.ac.jp/csj/index.html

Preparation

Step 1: Install dependencies

$ pip install -r requirements.txt

$ pip install -r chrome/requirements.txt

Step 2: Splits CSJ's wav files per sentence

$ ./split_wav.py --ignore_tag --ignore_kansuuji --ignore_anonymization --core /Volumes/Untitled/

Evaluation

Create recognized results from wav file by speech recognition system

In this repository, webkitSpeechRecognition in the Google Chrome is used as a baseline.

At first, run the Web application server as the following:

$  ./chrome/server.py

Next, open http://127.0.0.1:5000/ by Google Chrome.

Futhermore, connect a USB audio interface to your computer.

Finally, play wav files as the following:

$ ./play_wav.py --interval 2.0

After executing above commands, chrome/chrome_result.txt is created.

Calculate Word-Error-Rate (WER)

$ ./wer.py chrome/chrome_result.txt