How to process and inference?

[Linghuxc]

Hi, I used the pre-trained model for reasoning and found that mfa_model.zip and mfa_dict.txt were missing. I downloaded the relevant models from the official mfa and created folders by myself to put them in.

However, the output shows:

Do I need to perform the data processing part first? After entering the following command： show:

How should I solve this problem and I need help with!

[Zain-Jiang]

I'm sorry that I made mistakes in recent updates. Thank you for pointing out this problem. I have pushed a new readme file to guide the preprocess process and updated the corresponding codes.

[Linghuxc]

Hi, I am trying to train fluentspeech with LibriTTS,.Could you tell me what subset of LibriTTS do you use?

[Zain-Jiang]

I used train-clean-100 + train-clean-300.

[Linghuxc]

Maybe train clean 100 and train clean 360? Do you train one subset first and then train the next? Or do you mix two subsets together for training?

[Zain-Jiang]

yes, train clean 100 and train clean 360. I just mixed two subsets together for training in the experiments of our paper. But in my later experiments, mixing three subsets together shows better zero-shot capabilities.

[Linghuxc]

I have now pre-processed train_clean_100. Should I mix the two subsets again and preprocess them again or can I just process train_clean_360 on the current basis?

[Zain-Jiang]

The code does not support dataset concatenation. Sorry that you need to mix the two subsets again and preprocess them again.

[Linghuxc]

Ok, thanks for your reply!

[Zain-Jiang]

You're welcome! If you have any more questions or need further assistance, feel free to ask.

[Linghuxc]

Hi, I'm training the model on libritts and vctk, Use this command CUDA_VISIBLE_DEVICES=0 python tasks/run.py --config egs/spec_denoiser.yaml --exp_name spec_denoiser --reset I have a question.

I trained vctk using egs/spec_denoiser.yaml, but should I train libritts using egs/spec_denoiser.yaml or egs/spec_denoiser_libritts.yaml?

In addition, I have a friendly reminder, you can update the readme, python data_gen/tts/run_mfa_train_aligh.shmay need to change h to n, and then use bash to execute.

[Zain-Jiang]

I'm sorry that the egs/spec_denoiser_libritts.yaml is out-dated. You can use directly use egs/spec_denoiser.yaml to train the model on libritts.

Thanks for your advice! I will update the readme.