> Complete local text-to-waveform speech synthesis under 10M parameters.
In case, like me, you hoped "complete" voice might mean both stt and tts. Not to speak poorly of it, just clarifying.
> English only, with one fixed male voice. This is not zero-shot voice cloning.
(And then a bunch of statements on limitations that I read as 'quality can be spotty but if you play with it it should be fine') But like. In <10M params I'm not judging:)
semiquaver 4 days ago [-]
When would “text-to-waveform speech synthesis” ever imply speech to text?
yjftsjthsd-h 4 days ago [-]
The HN title is "Inflect-Micro-v2: complete voice in 9.36M parameters".
tmaly 4 days ago [-]
This is impressive. I wish there were a voice clone option.
fastball 4 days ago [-]
With so few parameters, I imagine a voice fine-tune might be readily tractable.
4 days ago [-]
NetOpWibby 4 days ago [-]
The inflections are weird but this doesn't sound like a robot. Not bad!
billdueber 4 days ago [-]
I keep seeing tts stories here. Is it just an interesting subset of the llm world, or is there a huge use case I’m somehow missing?
SamPatt 3 days ago [-]
I just built my own voice assistant with my Pebble Time 2 watch and it uses a VPS hosted TTS (Piper) and Hermes agent.
I learned about all of these projects on HN at one point or another.
eightysixfour 4 days ago [-]
I use STT/TTS to interface with a local LLM for Home Assistant in my house.
_davide_ 4 days ago [-]
I'm do so as well, i tried qwen3 omni 3 but it was ridiculously stupid, and i ended up with stt thinker and tts. kokoro for now
K0balt 4 days ago [-]
How heavy in inference on this? The model would easily fit on many microcontroller modules, I wonder if they could run it?
sudb 4 days ago [-]
this is extremely encouraging for individuals/small companies being able to train pareto-frontier TTS models (specifically compute required to run vs quality of model output)
StilesCrisis 4 days ago [-]
I'd love to hear it but it seems your quota is exhausted.
I think we need more neurons in the human brain than parameters in this model for speech. I wonder what it says about the human brain vs LLM efficiency.
jsomedon 4 days ago [-]
amazing quality for such small size!
itake 4 days ago [-]
Amazing quality for small size, but definitely not that enjoyable to listen to.
IMHO, its at about the same quality level of historic TTS tools.
stavros 4 days ago [-]
I'm not sure which historic tools you mean, but to me this sounds much better than anything older than ten years ago.
itake 4 days ago [-]
I compared the macos Samantha just now and I guess the inflect-micro is marginally better...
leobg 4 days ago [-]
Ivona „Joey“, „Amy“
phoenixranger 4 days ago [-]
amazing! was looking for something similar
mcbetz 4 days ago [-]
Alternative title: Text to speech in 9.36M, English only.
fintuner 4 days ago [-]
[flagged]
zenith605 4 days ago [-]
[flagged]
amelius 4 days ago [-]
[dead]
afdsaifdoi 4 days ago [-]
[dead]
Rendered at 09:07:17 GMT+0000 (Coordinated Universal Time) with Vercel.
here my implementation with speech dispatcher and server: https://github.com/skorotkiewicz/inflect-speechd
thanks for shearing!
> Complete local text-to-waveform speech synthesis under 10M parameters.
In case, like me, you hoped "complete" voice might mean both stt and tts. Not to speak poorly of it, just clarifying.
> English only, with one fixed male voice. This is not zero-shot voice cloning.
(And then a bunch of statements on limitations that I read as 'quality can be spotty but if you play with it it should be fine') But like. In <10M params I'm not judging:)
I learned about all of these projects on HN at one point or another.
https://inflect-tts.geronimo-labs.com
code: https://github.com/geronimi73/inflect-tts
On my iPhone 14 Pro the page crashes after 2-3 plays. I wonder if it uses too much memory?
defaulting to wasm on ios devices now
Disclaimer: I am the author
IMHO, its at about the same quality level of historic TTS tools.