Some letters sound not unclear or not fully pronounced
Hello again! As you are aware, I've been using this model to generate speech for Belarusian Anki Vocab cards. One thing I am noticing to my non-native ear is that some letter sounds are not coming through clearly or sound "swallowed" (if that makes sense). For example, I seem to notice this the most with Ц and Ць. Is there a good simple way to improve this in the model?
Hello! Is there a chance that you didn't use the phonemizer? The model has a readme with an instruction now
Oh, that's possible! Let me try that and I'll get back to you. Thanks!
Nope, I'm basically cloning your space (and using the same model) to generate the speech. For example, I'm able to generate the same issue when I use супраць as an input word for the instance running on your space. The difference seems to be more noticeable when you listen to the recording on Wiktionary: https://en.wiktionary.org/wiki/%D1%81%D1%83%D0%BF%D1%80%D0%B0%D1%86%D1%8C. Maybe this is just me and my untrained ear not picking up on the similarity. I can hear the correct sound on Wiktionary yet the synthesized speech sounds like ть. With that said, this is appears to be more of a me problem than a problem with the TTS code/model.