Complete Guide for fine tunning ( from custrom dataset preparation to fine tuneing over that data ) for Single speaker-no english speaker

[SanketDhuri]

can some give complete and step by process to train this model as answers for following questions:

1. how to prepare custom dataset for fine tuning a single speaker ?
2. what are hyper parameters as best to ta finetune and inference no-English / Devnagari language like Marathi ?

[slooi]

Hey @SanketDhuri did you ever learn the steps to making a custom dataset? Currently I'm attempting to train my own voice model and was looking for a streamlined guide

[martinambrus]

depending on the dataset language needed, if it's English, you can try and check my preparation notebooks here

[SanketDhuri]

@slooi I followed to the same guidelines provided here. I used audio editing tools to manually create custom data. I had to prepare the transcripts manually because I couldn’t find a suitable model that could simplify the transcription process. This was mainly because I was training a language that wasn’t supported by any of the available ASR models.

If you’re facing any hardware limitations, I recommend using Kaggle. I trained both stages on Kaggle.