+- Like Ra's Naughty Forum (https://www.likera.com/forum/mybb)
+-- Forum: Fetishes, obsessions, traits, features, peculiarities (https://www.likera.com/forum/mybb/Forum-Fetishes-obsessions-traits-features-peculiarities)
+--- Forum: Kinky Artificial Intelligence (https://www.likera.com/forum/mybb/Forum-Kinky-Artificial-Intelligence)
+---- Forum: Voice AI (https://www.likera.com/forum/mybb/Forum-Voice-AI)
+---- Thread: AllTalk TTS (/Thread-AllTalk-TTS)
RE: AllTalk TTS - Like Ra - 28 Feb 2025
Anyway, how do you use F5? Their website? Locally? If locally, how did you install it? What UI?
RE: AllTalk TTS - brandynette - 29 Apr 2025
(15 Feb 2025, 00:02 )Like Ra Wrote: AllTalk TTS
https://github.com/erew123/alltalk_tts/tree/alltalkbeta
AllTalk V2 Core Functionality
Comprehensive setup utilities for Windows & Linux
Multiple TTS engine support
- Coqui XTTS TTS
- Coqui VITS TTS
- Piper TTS
- Parler TTS
- F5 TTS
- Other TTS engines can be coded in
Can be used together with https://www.likera.com/forum/mybb/showthread.php?tid=4314
in my last year of development 6 struggling with this in particular
COQUI XTTS is dead. its a dead project & stuck before optimization while being very unstable. API server glitches, fails generations, has timing errors & all kinds of bhild issues.
COQUI VITS
same i ssues.
Piper. got it to work, quality, simplicity & flexibility is bad 6 i want to easily get it running. was too clunky & complex so i decided to leave it
Parler Very good model. Very heavy model. its for rish kids giggle
F5 super hard AF cuss documentation is tribal. on has to dig in post of posts & post in git foleders under git folders. in my opinion it was a hackaton built as prof of concept
it mutated into F5-TTS. which sucked balls as it consumed all my 8GB of ram & then devilered terrible speeds wiht no valuabel quality gain.
Now im using a perfect solution for my current hardware limitations
KORORO-FastAPI a minified, optimized for speed & very low resource usage with the actual ability to run on the CPU. absolutely stable. it has never crashed.
so it builds the speech model loading & running infrastructure for Kokoro & file handling. where an URL encoded string post to api gets processed & output as an MP3 & weblink so only needs an universal html audio tags src field
has a good list of fixated voices who have clear speech with very little mistakes or weird sounds. & zthe ability to mix them up just by using voice names & their wanted percentage. works well with 2, 3 get glitchy.
It has good community, dev team is commited 6 has done an amazing job well solving my issue perfectly
hugginface
https://huggingface.co/hexgrad/Kokoro-82M
github
https://github.com/remsky/Kokoro-FastAPI
tis so mindles and simple this is all you need to run to boot kokoro up.
docker run --cpus=4 --memory=32g --restart always -p 8880:8880 ghcr.io/remsky/kokoro-fastapi-cpu:latest
DELETE AFTER YOU SAW IT
2025-04-29_14-34-30.mp4 (Size: 35.28 MB )
(27 Feb 2025, 00:43 )SensualTiffany Wrote:(26 Feb 2025, 12:36 )Like Ra Wrote:(26 Feb 2025, 05:21 )SensualTiffany Wrote: Did you do this online or did you load the model in your computer?As the thread title suggests - locally installed AllTalk TTS v2 and locally trained Nikki's voice.
(26 Feb 2025, 05:21 )SensualTiffany Wrote: PM me the script and I’ll work on it.It's the second one in this post: https://www.likera.com/forum/mybb/Thread-NSFW-AI-ChatBots?pid=79132#pid79132
So, I didn't spend a huge amount of time and it took a bit of tweaking, but here is the beginning of your text, generated from the T5-TTS. I added the Theta because there were interestingly different silences, created by the TTS. I like Theta or Background music myself.
[attachment=65995]
Couldn't help myself: Here is a version with some...added sound.
[attachment=65996]
Cheers.
(24 Feb 2025, 22:48 )SensualTiffany Wrote:(24 Feb 2025, 14:05 )Helga Wrote:(23 Feb 2025, 22:42 )SensualTiffany Wrote: Well, I just had to do a little work.Damn gurl, you are spirit talker, with such skills you can bring back the classics or even more.
Here is TTS Nikki with a short, barely edited 4 minute clip of a section of script.
Enjoy.
Beginning are a series of just WORDS or PHRASES until the file starts talking about Pink Mist.
The program uses only a 15min sample and creates the TTS from it. I played around with several base samples, so the difference in tones.
I’ll keep playing with it, but the initial plan is to modify one of my favorite Nikki files. All those disturbing “Boy” references will be more feminine friendly.
Amzing werk, intonation and sound is quite well.
For 5 min test run, it is outstanding A+
Can you try Madame Y robotic voice for sake of testing ? on the other hand, nevermind.
You aeady did "magic", I suppose if AI can do Nikki, Madame Y is weak candidate to challange the system.
Your wish is my command: A little Madame Y.
Had to strip out music first, but I think it might be passable when mixed with a binaural and some decent music.
Thoughts?
Plain: [attachment=65981]
With Theta: [attachment=65982]
With Music: [attachment=65983]
@SensualTiffanyhow did you generate the tones? im absolutly interested in your effects