Add Speech-to-speech
in progress
spanishmillennial
Shouldnt this be marked as Done now?
m
moshe
Actually, I must correct my previous observation. The model cannot handle high emotion. It renders a flat, low-energy revoicing. It also leaks the speaker's accent very badly into the revoicing, while simultaneously failing to apply the vocal sub-tones that make the AI voice sound alive.
The thing can render normal conversational dialogue quite nicely, but it falls completely flat on its face when attempting to render, say, a rousing pre-battle speech. Using it wins us nothing over the text-to-speech regeneration, in such cases. In fact, text-to-speech is superior.
m
moshe
The version released today is quite good. I especially like the ability to re-voice all the non-word verbals, e.g. laughter, sneers, etc. That makes the dialogue really come alive.
Still, the model could use some improvement. I notice that it brings some of my accent along for the ride. It is subtle, but noticeable to a careful listener. It would be good if we could provide the written text and the audio simultaneously, with functionality to line up the two, so that the AI can replicate the emotion and put stress on the correct syllables, but avoid, for example, screwing up the consonants. Also, for the love of all things literary -- give us a multilingual version of this thing!
G
Greg Baker
I have a very serious use case for this. I want accent reduction on videos. I'd estimate that I have about 10,000 hours of lecture audio per month that I could use this with.
m
moshe
When will this be released? The damned AI is completely incapable of maintaining emotions, diction, pacing -- anything at all -- over the course of an impassioned scene. It can barely do rage across a single sentence, never mind a paragraph! I am fighting with the thing over every damned sentence!
S
Shivam Rastogi
Rayan when is this planned to land?
CarcomCars
YESS! Oh my goodness...you guys are speaking my language. And yes PLEASE NOT JUST BY MICROPHONE!!!! LET US ATTACH FILES.
S
Shane Zammit
can't wait for this
Sebastian Plasschaert
Great idea. But please not just by microphone. Would be great if we can upload speech-files.
Rayan
in progress
Load More
→