Dataset

#4
by rahul7star - opened

I will try this soon but
Can i use dataset which just has snac codes only for training
Or we feed audio data set which will be best bet from these 2 dataset

rahul7star/hindi-speech-dataset

Or

With nano layers
rahul7star/vaani-snac-cleaned

@rahul7star You unfortunately have to either

  1. Convert snac codes back into audio and then convert to bicodec codes(MiraTTS uses that).
  2. Directly convert audio into bicodec codes. The notebook has a process audio section which does this.

So probably use rahul7star/hindi-speech-dataset

Also you said MiraTTS is a fine tune of spark but i dont see the mira listing under spark fine tunes , spark has https://huggingface.co/models?other=base_model:finetune:SparkAudio/Spark-TTS-0.5B which one is yours

does your model works with hindi ? I trained on hindi data set seems not working
space - https://huggingface.co/spaces/rahul7star/Mira-TTS.
model - rahul7star/mir-TTS

Sign up or log in to comment