Dataset suitable for german Parler TTS model training (via DataSpeech)?

#2
by Thorsten-Voice - opened

Hello 👋,

as this is my first HuggingFace dataset upload i am asking for some review before i continue with the next steps.
My goal is to build an open source parler tts model based on my Thorsten-Voice audio recordings ( @ylacombe ).

After my upload i can see a preview and the dataset viewer, which i like 😊. But there seems not be be any filter in the viewer table header, like here e.g.: https://huggingface.co/datasets/mythicinfinity/libritts_r

ViewerHeaderLibri.png

Is this because of a problem with my dataset or can this features be added by configuration?

Next, i started with DataSpeech and ran following command:

python main.py "Thorsten-Voice/TV-44kHz-Full" --configuration "all" --output_dir ./Thorsten-Voice/tmp_tv_full/ --text_column_name "text" --audio_column_name "audio" --cpu_num_workers 8 --rename_column --repo_id "Thorsten-Voice/TV-44kHz-DataSpeech" --apply_squim_quality_estimation

This created this repo: https://huggingface.co/datasets/Thorsten-Voice/TV-44kHz-DataSpeech

As it's primarily meant for english i guess i have to adjust or replace this g2p with a german compatible logic, am i?
https://github.com/huggingface/dataspeech/blob/aecbcfde51bfcada69b1983ae46b24bf2fe545a0/dataspeech/cpu_enrichments/rate.py#L3

Long story short - am i on the right track to create an open source german parler tts model?

Sorry if this is the wrong place to ask. If there's a better way to discuss (on GitHub) or on the DataSpeech repo please let me know. Thanks for any feedback 😊.

cc @albertvillanova @lhoestq @severo .

Sign up or log in to comment