Retrieval-based-Voice-Conversion-WebUI
Retrieval-based-Voice-Conversion-WebUI copied to clipboard
Certain sounds are consistently mispronounced with English, especially when first speaking
Hello!
I've noticed that there's some words that never pronounce correctly when speaking English. This becomes exasperated when these sounds are spoken early in a statement. This is consistent across all models and all types of configuration. It seems to be an issue with RVC itself, as I don't run into this problem with other model types such as Beatrice.
One easy example is the word "coral". The correct english pronunciation can be heard here: https://drive.google.com/file/d/1-2GILMz7-UbN1NDcaZ7JFaG8qPHcKaPO/view
But the pronunciation when I say "coral" using voice changer and RVC models (regardless of chunk num or extra data length) can be heard here: https://drive.google.com/file/d/1jskEbUkE_hhZII3fHmKtGi-tAvtkRt3l/view?usp=sharing
It sounds really off! This is one of the few problems I run into when using real time voice changing, as I often have to keep in mind the words that cause problems and avoid saying them. It seems to happen a lot with 'c' sounds and sounds such as the 'ph/f' sound. It becomes less pronounced if it's spoken later in a sentence. For example "Coral is nice" will cause the issue badly, but if I were to say "I like to see the coral" it would likely pronounce correctly.
I'm positive it's not a configuration issue as I've verified others have the same problem, and I've attempted to tweak every option to see if it's something possible to fix.
Interestingly, the same issue seems to happen using RVC's file conversion, so it's not limited to real-time.
@Mojobones did you find a solution to this?
@Mojobones did you find a solution to this?
I didn't, but it's not as much of an issue these days. I can trigger if it I specifically try, but for the most part seems to not be as much of a problem.
This issue was closed because it has been inactive for 15 days since being marked as stale.