alt Hacker NewsPocket TTS: A high quality TTS that gives your CPU a voice
Comments
The speed of improvement of tts models reminds me of early days of Stable Diffusion. Can't wait until I can generate audiobooks without infinite pain. If I was an investor I'd short Audible.
I read this, then realized I needed a browser extension to read my long case study and made a browser interface of this and put this together:
Nice!
Just made it an MCP server so claude can tell me when it's done with something :)
Oh this is sweet, thanks for sharing! I've been a huge fan of Kokoro and event setup my own fully-local voice assistant [1]. Will definitely give Pocket TTS a go!
How feasible would it be to build this project into a small static binary that could be distributed? The dependencies are pretty big.
Love this.
It says MIT license but then readme has a separate section on prohibited use that maybe adds restrictions to make it nonfree? Not sure the legal implications here.
Eep.
So, on my M1 mac, did `uvx pocket-tts serve`. Plugged in
> It was the best of times, it was the worst of times, it was the age of wisdom, it was the age of foolishness, it was the epoch of belief, it was the epoch of incredulity, it was the season of Light, it was the season of Darkness, it was the spring of hope, it was the winter of despair, we had everything before us, we had nothing before us, we were all going direct to Heaven, we were all going direct the other way—in short, the period was so far like the present period, that some of its noisiest authorities insisted on its being received, for good or for evil, in the superlative degree of comparison only
(Beginning of Tale of Two Cities)
but the problem is Javert skips over parts of sentences! Eg, it starts:
> "It was the best of times, it was the worst of times, it was the age of wisdom, it was the epoch of belief, it was the epoch of incredulity, it was the season of Light, it was the spring of hope, it was the winter of despair, we had everything before us, ..."
Notice how it skips over "it was the age of foolishness,", "it was the winter of despair,"
Which... Doesn't exactly inspire faith in a TTS system.
(Marius seems better; posted https://github.com/kyutai-labs/pocket-tts/issues/38)
It's pretty good. And for once, a software-engineering-ly high-quality codebase, too!
All too often, new models' codebases are just a dump of code that installs half the universe in dependencies for no reason, etc.
Perhaps I have been not talking to voice models that much or the chatgpt voice always felt weird and off because I was thinking it goes to a cloud server and everything but from Pocket TTS I discovered unmute.sh which is open source and I think is from the same company as Pocket TTS/can I think use Pocket TTS as well
I saw some agentic models at 4B or similar which can punch above its weights or even some basic models. I can definitely see them in the context of home lab without costing too much money.
I think atleast unmute.sh is similar/competed with chatgpt's voice model. It's crazy how good and (effective) open source models are from top to bottom. There's basically just about anything for almost everyone.
I feel like the only true moat might exist in coding models. Some are pretty good but its the only industry where people might pay 10x-20x more for the best (minimax/z.ai subscription fees vs claude code)
It will be interesting to see if we will see another deepseek moment in AI which might beat claude sonnet or similar. I think Deepseek has deepseek 4 so it will be interesting to see how/if it can beat sonnet
(Sorry for going offtopic)
Good quality but unfortunately it is single language English only.
Just added it to my codex plugin that reads summary of what it finishes after each turn and I am spooked! runs well on my macbook, much better than Samantha!
I'm missing the old days that connecting a SPOKE256 to the Spectrum and making it speak, looked like magic.
This is amazing. The audio feels very natural and it's fairly good at handling complext text to speech tasks. I've been working on WithAudio (https://with.audio). Currently it only uses Kokoros. I need to test this a bit more but I might actually add it to the app. It's too good to be ignored.
it'd be nice to get some idea of what kind of hardware a laptop needs to be able to run this voice model.
It'd be great if it supported stdin&stdout for text and wav. Then it could get piped right into afplay
I'm sure I'm being stupid, but every voice except "alba" I recognize from Les Miserables; is there a character I'm forgetting?
Would be nice if preview supports variable speed.
Is there something similar for STT? I’m using whisper distill models and they work ok. Sometimes it gets what I say completely wrong.
It's cool how lightweight it is. Recently added support to Vision Agents for Pocket. https://github.com/GetStream/Vision-Agents/tree/main/plugins...
I love that everyone is making their own TTS model as they are not as expensive as many other models to train. Also there are plenty of different architecture.
Another recent example: https://github.com/supertone-inc/supertonic
Perfect timing that is exactly what I am looking for for a fun little thing I'm working on. The voices sound good!
It's very impressive! I'm mean, it's better than other <200M TTS models I encounter.
In English, it's perfect and it's so funny in others languages. It sounds exactly like someone who actually doesn't speak the language, but got it anyway.
I don't know why Fantine is just better than the others in others languages. Javer seems to be the worst.
Try Jean in Spanish « ¡Es lo suficientemente pequeño como para caber en tu bolsillo! » sound a lot like they don't understand the language.
Or Azelma in French « C'est suffisament petit pour tenir dans ta poche. » is very good.I mean half of the words are from a Québécois accent, half French one but hey, it's correct French.
Però non capisce l'italiano.
voices sound great! i see sample rate can be adjusted, is there any way to adjust the actual speed of the voice?
Haven't we had TTS for like 20+ years? Why does AI need to be shoved into it all of a sudden. Total waste of electricity.
>If you want access to the model with voice cloning, go to https://huggingface.co/kyutai/pocket-tts and accept the terms, then make sure you're logged in locally with `uvx hf auth login` lol
Relative to AmigaOS translator.device + narrator.device, this sure seems bloated.
I'm psyched to see so much interest in my post about Kyutai's latest model! I'm working on part of a related team in Paris that's building off Kutai's research to provide enterprise-grade voice solutions. If anyone building in this space I'd love to chat and share some our upcoming models and capabilities that I am told are SOTA. Please don't hesitate to ping me via the address in my profile.