Subscribe
Sign in
General
p/generalShare and discuss tech, products, business, startups, or product recommendations
Start new thread
Abhinav Yadav

What are biggest limitations of AI generated voices you have faced?

by•

My team at Wavel has been working on building and improving AI generated voices from last few months. Would love to get opinions from the community about the product or algorithms they have used in recent times.

Add a comment

Replies

Best
Michael Choupak
it's high time that AI voices pass their own version of the Voice Turing Test to sound just like humans. At the moment, they all have a bit of a robotic vibe to them. which text-to-speech tool do you think stands out?
Abhinav Yadav
@michael_choupak This is what we have been working on for a while. Removing the robotic aspect of it. Since I am developing Wavel I will be biased for it. However, given what was offered 6 months ago there is a drastic improvement. In my opinion, the outcome also depends on the content you are using it for. For content like explainer videos, the AI is at par.
Simon Peter Damian
Launching soon!
Most can't produce accented sounds very well. They are mostly in western voices
Abhinav Yadav
@theterminalguy Interesting. Have you tried voice cloning to solve this problem?
Simon Peter Damian
Launching soon!
@abhinav_wavel not at all. I did read Facebook's. Voice box paper some months back which seem interesting but I'm yet to try out cloning
Olaf
Do you clone voices as well?
Isao Fukata
I have been using Amazon Polly, and I feel that the English generated voice has become much more natural this year. However, the intonation of other languages is still unnatural. In particular, the intonation of exclamations is often strange.
Daniel Burns
We've been using Speechelo for the voice generator (we've been using it for English only) for quite some time now, as it has proven to be the best thus far for us, however, it has some drawbacks. Sometimes the pronunciation is off and no matter how many times the text is altered, you can still hear the unnatural (robotic) accent.