search

BenAAndrew / Voice-Cloning-App

A Python/Pytorch app for easily synthesising human voices
BSD 3-Clause "New" or "Revised" License
1.4k stars 233 forks source link

Alphabet file does not seem to be used in Google Colab #94

Closed Ca-ressemble-a-du-fake closed 2 years ago

Ca-ressemble-a-du-fake commented 2 years ago

Hi,

Great project ! I post my comment here as an issue instead of there as a comment because it deals with the Colab training page specifically.

So on the Colab training page, I followed the instructions regarding the alphabet file, and I created a "French.txt" file in Voice-Cloning/alphabets folder based on German.txt alphabet. The alphabet is found because it is selected in the drop-down list but then I get the "Invalid characters in text" error even though the characters listed in the error are present in "French.txt" alphabet file.

To make the training work I had to clone this repo, modify DEFAULT_ALPHABET in training's __init__ module (code changes made directly in Colab were not taken into account). Consequently it seems that the custom made alphabet file is not used or I made something wrong.

BenAAndrew commented 2 years ago

Good spot, I'll get that fixed now

BenAAndrew commented 2 years ago

That should be fixed now, please reopen if not. (P.S. You can download alphabet files from https://coqui.ai/ rather than creating yourself)

Ca-ressemble-a-du-fake commented 2 years ago

I'll check it out thanks for the fix! Regarding alphabet file, Coqui adds also very weird characters like ("ờ‐―₽→∆ɔɛ") so I don't feel so confident using them (I don't know if these added chars will bloat the memory consumption). Anyway thanks for the tip.