resemble-ai/chatterbox

Wrong numbers pronouncation in Russian language

Opened this issue · 1 comments

If text prompt contains numbers or dates written as digits (1, 1900), they are not prounounced in Russian, even though the language is specified.
Example:
text: 25470
language_id = "ru"
Reference voice is Russian
Below are the examples of chatterbox generated audio and google generated audio with correct pronouncation.

Chatterbox Audio

Google Audio(Correct)

The speech synthesis task does not guarantee the conversion of numbers and abbreviations into correct Russian. Raw text cannot be fed to the generator without pre-processing, and in this case, Russian normalization is required. For Example https://github.com/saarus72/text_normalization