Inflect-Micro-v2: complete voice in 9.36M parameters

(huggingface.co)

212 points | by nateb2022 3 days ago ago

34 comments

  • modinfo 3 days ago ago

    This is amazing, the quality blow my mind for such small model! I just replaced my old onnx model with yours!

    here my implementation with speech dispatcher and server: https://github.com/skorotkiewicz/inflect-speechd

    thanks for shearing!

  • yjftsjthsd-h 3 days ago ago

    Couple highlights:

    > 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 2 days ago ago

      When would “text-to-waveform speech synthesis” ever imply speech to text?

      • yjftsjthsd-h 2 days ago ago

        The HN title is "Inflect-Micro-v2: complete voice in 9.36M parameters".

  • NetOpWibby 2 days ago ago

    The inflections are weird but this doesn't sound like a robot. Not bad!

  • tmaly 3 days ago ago

    This is impressive. I wish there were a voice clone option.

    • fastball 3 days ago ago

      With so few parameters, I imagine a voice fine-tune might be readily tractable.

  • billdueber 2 days ago 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 2 days ago 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 2 days ago ago

      I use STT/TTS to interface with a local LLM for Home Assistant in my house.

      • _davide_ 2 days ago 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 2 days ago ago

    How heavy in inference on this? The model would easily fit on many microcontroller modules, I wonder if they could run it?

  • sudb 2 days ago 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)

  • da-x 2 days ago ago

    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.

  • StilesCrisis 2 days ago ago

    I'd love to hear it but it seems your quota is exhausted.

  • jsomedon 3 days ago ago

    amazing quality for such small size!

  • itake 3 days ago 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 2 days ago 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 2 days ago ago

        I compared the macos Samantha just now and I guess the inflect-micro is marginally better...

      • leobg 2 days ago ago

        Ivona „Joey“, „Amy“

  • phoenixranger 2 days ago ago

    amazing! was looking for something similar

  • mcbetz 2 days ago ago

    Alternative title: Text to speech in 9.36M, English only.

  • fintuner 2 days ago ago

    [flagged]

  • zenith605 2 days ago ago

    [flagged]

  • amelius 2 days ago ago

    [dead]

  • afdsaifdoi 2 days ago ago

    [dead]