There are so many models available in coqui, Can someone point me the best model out of them, which gave the best results so far.

[Rajashekhar-Reddy]

I'm too confused about choosing the perfect modal.
Appreciate it, if anyone could add the output results here for the suggested modal.

[imjonos]

Same question)

[thorstenMueller]

Hi @Rajashekhar-Reddy and @imjonos. "Best" and "perfect" models are very subjective and individual, but maybe you can get an idea by listening to them in my audio sample comparison videos.

• https://youtu.be/HojuVmW5LUI
• https://youtu.be/Vnjv2L31eyQ

[imjonos]

Thanks 👍

[pivolan]

if you want train new models, at the current time the best results for me when training from the scratch on small datasets 3-5 hours, best results I got from vits model single voice. Also if you want finetune your own voice after opensource model, there are also two experiments was on my side:

1. finetune single voice - got much more clear audio voice, but create errors in prnounciations, minimum dataset for finetune 1hours, good results will be from 4 hours, without pronounce errors.
2. finetune using yourtts with speaker encoder d_vector, usefull even you have only 15 minutes, and finetune it will be responsible, and good enough quality. If you have 1 hours voice not so clear as in single voice, but much less errors in prnounciation. And for 4 hours I didn't find difference in pronounce errors, but in single voice model audio much more clear.

[waqasawan9996]

which model contain that polly voices specifically matthew

[eginhard]

Polly is Amazon's commercial TTS offering: https://docs.aws.amazon.com/polly/latest/dg/API_Voice.html

[waqasawan9996]

hmm i was looking for a personal use that should sounds like matthew is there any pretrained models for coqui that should good in storytellig