How to get f0_state.pth, I don't understand

[xiaoyuTTS]

How to get f0_state.pth, I don't understand

[djkhz]

@xiaoyuTTS
Your work is incredible!! If you don't mind, I have a lot of questions concerning your work. Is there any way to contact you?
Is it possible for you to publish a full article and tutorial regarding your work?

[adiyoss]

Hi @xiaoyuTTS,
You should train an F0 quantizer model, see: https://github.com/facebookresearch/speech-resynthesis#f0-quantizer-model

[adiyoss]

Hi @djkhz,
All the details concerning this work here are listed in our manuscript here: https://arxiv.org/pdf/2104.00355.pdf

[Rongjiehuang]

Hi @adiyoss, It seems that f0_state.pth is needed before training the F0 quantizer model.

[sophia1488]

I also encountered this problem, but I found a package that could get f0_mean & f0_std.

features1=prosody.extract_features_file(file_audio)
features[:, 0] # mean
features[:, 1] # std

[adiyoss]

Hi @Rongjiehuang,
Yes you need to get the statistics for the f0, there is an example for such stats under datasets/LJSpeech and datasets/VCTK

[duyichao]

Hi @adiyoss, the code base provides f0_stats.pth for VCTK/LJSpeech (English only). If I want to get a f0_stats.pth for another language, how do I do that?

[speechless-z]

Hi @adiyoss, the code base provides f0_stats.pth for VCTK/LJSpeech (English only). If I want to get a f0_stats.pth for another language, how do I do that?

Hello, have you resolved the issue?

[PrabhjotKaurGosal]

Also interested in figuring out how to get the f0_stats.pth for a custom (Non English) dataset.