gheyret / uyghur-asr-ctc

Speech Recognition for Uyghur using deep learning

Speech Recognition for Uyghur using deep learning

Training:

this model using CTC loss for training.

Download pretrained model and dataset.

unzip results.7z and thuyg20_data.7z to the same folder where python source files located. then run:

python train.py

Recognition:

for recognition download only pretrained model(results.7z). then run:

python tonu.py test1.wav 

result will be:

        Model loaded: results/UModel_last.pth
            Best CER: 7.21%
             Trained: 473 epochs
The model has 26,389,282 trainable parameters

======================
Recognizing file .\test2.wav
test2.wav -> bu öy eslide xotunining xush tebessumi oghlining omaq külküsi bilen güzel idi

This project using

A free Uyghur speech database Released by CSLT@Tsinghua University & Xinjiang University

uyghur-asr-ctc's Issues

text

How did you turn the text in the original data set into standard Latin Uighur?

To repeat your result

Gheyret,
Thanks for your work, I tried to reproduce your work, but after 300 epochs (batch size = 20), CER was still around 10%, not sure how many epochs for you to reach 7.21%
One more thing I noticed that your dataset is a little different with the original dataset (openslr) --- for some utterances, for example, F178_007.wav, you have F178_007_1.wav and F178_007_2.wav. Can you elaborate it a little on the reason for you to do that modification?
Thanks for your answer!