It is not the best TTS but it is freaking amazing it can be done by such a small model and it is good enough for so many use cases.
Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model
81–90 of 383 posts
Re: Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model
#82Earlier quoted context omitted.
A robotic, explicitly non-natural voice would be perfectly acceptable, and even desirable, in many situations[...]it'd be enough is the speech if easy to recognize. We've had formant synths for several decades, and they're perfectly understandable and require a tiny amount of computing power, but people tend not to want to listen to them: https://en.wikipedia.org/wiki/Software_Automatic_Mouth https://simulationcorner…
Well, this one is a bit too jarring to the ears.
But there is no latency, as opposed to KittenTTS, so it certainly has its applications too.
Re: Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model
#83Earlier quoted context omitted.
[flagged]
Not open source. "You will need internet connectivity to validate your AccessKey with Picovoice license servers ... If you wish to increase your limits, you can purchase a subscription plan." https://github.com/Picovoice/orca#accesskey
Going online is a dealbreaker but if you really need it you could use ghidra to fix that. I had tried to find a conversion of their model to onnx (making their proprietary pipeline useless) but failed.
Hopefully open source will render them irrelevant in the future.
Re: Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model
#84Web version: https://clowerweb.github.io/kitten-tts-web-demo/ It sounds ok, but impressive for the size.
I got an error when I tried the demo with 6 sentences, but it worked great when I reduced the text to 3 sentences. Is the length limit due to the model or just a limitation for the demo?
Perhaps a length limit? I tried this:
"This first Book proposes, first in brief, the whole Subject, Mans disobedience, and the loss thereupon of Paradise wherein he was plac't: Then touches the prime cause of his fall, the Serpent, or rather Satan in the Serpent; who revolting from God, and drawing to his side many Legions of Angels, was by the command of God driven out of Heaven with all his Crew into the great Deep."
It takes a while until it starts generating sound on my i7 cores but it kind of works.
This also works:
"blah. bleh. blih. bloh. blyh. bluh."
So I don't think it's a limit on punctuation. Voice quality is quite bad though, not as far from the old school C64 SAM (https://discordier.github.io/sam/) of the eighties as I expected.
Re: Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model
#85Earlier quoted context omitted.
A robotic, explicitly non-natural voice would be perfectly acceptable, and even desirable, in many situations[...]it'd be enough is the speech if easy to recognize. We've had formant synths for several decades, and they're perfectly understandable and require a tiny amount of computing power, but people tend not to want to listen to them: https://en.wikipedia.org/wiki/Software_Automatic_Mouth https://simulationcorner…
Well, this one is a bit too jarring to the ears.
Try this demo, which has more knobs:
Re: Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model
#86it would be great if there is typescript support in the future
Re: Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model
#87I don't mind so much the size in MB, the fact that it's pure CPU and the quality, what I do mind however is the latency. I hope it's fast. Aside: Are there any models for understanding voice to text, fully offline, without training? I will be very impressed when we will be able to have a conversation with an AI at a natural rate and not "probe, space, response"
Voice to text fully offline can be done with whisper. A few apps offer it for dictation or transcription.
Re: Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model
#88I don't mind so much the size in MB, the fact that it's pure CPU and the quality, what I do mind however is the latency. I hope it's fast. Aside: Are there any models for understanding voice to text, fully offline, without training? I will be very impressed when we will be able to have a conversation with an AI at a natural rate and not "probe, space, response"
"The brown fox jumps over the lazy dog.." Average duration per generation: 1.28 seconds Characters processed per second: 30.35 -- "Um" Average duration per generation: 0.22 seconds Characters processed per second: 9.23 -- "The brown fox jumps over the lazy dog.. The brown fox jumps over the lazy dog.." Average duration per generation: 2.25 seconds Characters processed per second: 35.04 -- processor : 0 vendor_id : Au…
Hmm that actually seems extremely slow, Piper can crank out a sentence almost instantly on a Pi 4 which is a like a sloth compared to that Ryzen and the speech quality seems about the same at first glance.
I suppose it would make sense if you want to include it on top of an LLM that's already occupying most of a GPU and this could run in the limited VRAM that's left.
Re: Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model
#89Good TTS feels like it is something that should be natively built into every consumer device. So the user can decide if they want to read or listen to the text at hand. I'm surprised that phone manufacturers do not include good TTS models in their browser APIs for example. So that websites can build good audio interfaces. I for one would love to build a text editor that the user can use completely via audio. Text inp…
On Mac OS you can "speak" a text in almost every app, using built in voice (like the Siri voice or some older voices). All offline, and even from the terminal with "say".
Re: Show HN: Kitten TTS – 25MB CPU-Only, Open-Source TTS Model
#90"please join our DISCORD!"...