veelion

remove Chinese charactor from profiles/ko, to make sure Chinese is not mistakenly be detected as Korean

2

In: langdetect.detect(u'就了快速大幅') Out: 'ko' But the string is definitely Chinese. The problem is that, there are so many Chinese character in profiles/ko So, I remove it using the script fix-ko.py

For many technical terms, e.g. "AI", "GitHub", phonemes converted by seq2se2 are not the right pronunciation, so a custom dict is necessary. The added method `load_custom_phonemes(self, file_path)` to G2p, reads...

Add 'run_batch' mode for GPU encoding and decoding with batch_size >= 1

1

This mode improves the throughput of websocket server. Test result: * hardware-1: Platinum 8358P CPU @ 2.60GHz 15 cores + 80G memory, A5000 * 1 + 24G memory * hardware-2:...

How to pretrain the model from scratch？

2

Hi @rajatsen91 Thanks for sharing the finetuning code, I have finetuned the 2.0 model with my own data and get the better result. I want to train the model from...

Hi all, I use the model to forecast use same data but different batch_size, and get different size. Firstly, batch_size=1, inference process: for i, input in enumerate(inputs): result, _ =...