OnceJune

Hi, I tried to create linguistic feature for mandarin follow this recipe: egs/mandarin_voice I noticed that in the question file, mandarin used only L-\*/C-\*/R-\* phones, while other locale used all...

[Bug] duration predictor safe guard

1

https://github.com/rishikksh20/LightSpeech/blob/d9290f755f02d33d520c2304c5b6624f87864e55/core/duration_modeling/length_regulator.py#L89 Runtime error when length predicted negative value. Maybe need to modify to: `out.append(x_.repeat(max(int(d_), 0), 1)) `

Duration in filelist starts with -2?

1

Hi, thanks for sharing the code. When I tried to run lightspeech on my own dataset, I found in sample file list, the sum duration always starts from -2: https://github.com/rishikksh20/LightSpeech/blob/d9290f755f02d33d520c2304c5b6624f87864e55/filelists/valid_filelist.txt#L1...

How about the audio quality?

6

Hi, thanks to the implement, the inference speed is impressive. How about the audio quality? And have you tried v2 config? Thanks in advance.

Has anyone tried fastspeech2s? can it reach the audio quality from fastspeech2+ganvocoder?

[Question] How to apply streaming inference

1

Hi, is fs2 only support synthesis whole sentence? How should I make it output chunk by chunk, e.g. 50 frames every time? Thanks in advance.

More language support?

6

Hi, I'd like to know will mistral planning to support more languages?

VITS2?

4

Hi, VITS2 has been released at:https://arxiv.org/pdf/2307.16430.pdf, do you have the plan to release the code?

无监督训练数据构建问题

3

https://github.com/vdogmcgee/SimCSE-Chinese-Pytorch/blob/e04263b1846fceef8dff60ead29d35e02947918b/simcse_unsup.py#L73 这里的实现是添加了自身两次，想请教一下论文当中的”同一个句子的不同 Dropout Mask“是怎么体现的呢？

Hi, I tried to run test on a MacBook with CPU but got error: /Users/me/anaconda3/envs/rad-nerf/lib/python3.6/site-packages/torch/autocast_mode.py:141: UserWarning: User provided device_type of 'cuda', but CUDA is not available. Disabling warnings.warn('User provided device_type...