History

TianYuan 22e5527c84 fix textnorm data and readme		2021-08-30 07:03:44 +00:00
..
data	fix textnorm data and readme	2021-08-30 07:03:44 +00:00
README.md	fix textnorm data and readme	2021-08-30 07:03:44 +00:00
get_g2p_data.py	restructure frontend example	2021-08-16 08:31:37 +00:00
get_textnorm_data.py	format	2021-08-17 09:54:07 +00:00
make_sclite.sh	add make_sclite.sh	2021-08-23 03:27:54 +00:00
run.sh	add make_sclite.sh	2021-08-23 03:27:54 +00:00
test_g2p.py	add sclite test for text_frontend	2021-08-21 08:35:46 +00:00
test_textnorm.py	add sclite test for text_frontend	2021-08-21 08:35:46 +00:00

Chinese Text Frontend Example

Here's an example for Chinese text frontend, including g2p and text normalization.

G2P

For g2p, we use BZNSYP's phone label as the ground truth and we delete silence tokens in labels and predicted phones.

You should Download BZNSYP from it's Official Website and extract it. Assume the path to the dataset is ~/datasets/BZNSYP.

We use WER as evaluation criterion.

For text normalization, the test data is data/textnorm_test_cases.txt, we use | as the separator of raw_data and normed_data.

We use CER as evaluation criterion.

If you want to use sclite to get more detail information of WER, you should run the command below to make sclite first.

./make_sclite.sh

Run the command below to get the results of test.

./run.sh

The avg WER of g2p is: 0.027495061517943988

The avg CER of text normalization is: 0.006388318503308237