Skip navigation
We built an under 1B speech-to-text model that outperforms both open and closed source models on…
0:00 / 0:52

We built an under 1B speech-to-text model that outperforms both open and closed source models on…

FluidVoice

@FluidVoiceApp4.2K followers

Visit on X
752 likesDownloadOriginal post

44,457 views

We built an under 1B speech-to-text model that outperforms both open and closed source models on accuracy. Introducing Fluid Speech Pro. 🚀(1/6)

35 Comments

Comments from the original post on X (Twitter)

  • FluidVoice's profile picture

    @FluidVoiceApp

    Fluid Speech Pro understands you even in noisy cafes, trains, and crowded spaces. High noise robustness in models under 1B parameters has been a challenge. We're changing that. (2/6)

    14 likes

  • FluidVoice's profile picture

    @FluidVoiceApp

    Fluid Speech Pro is our latest launch under the Fluid Speech family of models. Fluid Speech is a family of fine-tuned, quantized models, ranging from just 150 MB to 600 MB, all under 1B parameters. We've built this family to support STT on any sort of edge device that you have. (3/6)

    14 likes

  • FluidVoice's profile picture

    @FluidVoiceApp

    Every Fluid Speech model is built for handling real-world noise. Crowded cafes. Loud TVs. Busy streets. Even the tiny 150 MB model. (4/6)

    12 likes

  • FluidVoice's profile picture

    @FluidVoiceApp

    This is just the beginning of what is possible with local AI. We believe that privacy and intelligence can coexist. And we are on our way. This is our first step towards that. We also make dictation cleanup models that runs extremely fast, even on your older devices. (5/6)

    11 likes

  • FluidVoice's profile picture

    @FluidVoiceApp

    The entire Fluid Speech family is already live on the FluidVoice app on Mac, and will be available on the other OS's in the coming days! Give them a spin and tell us what you think! FluidVoice: (6/6)

    22 likes

  • FluidVoice's profile picture

    @FluidVoiceApp

    Benchmark numbers here:

    7 likes

  • Zain ul Abidin's profile picture

    @ZuA_1245

    This actually is a huge upgrade over the previous models. I have seen that the accuracy is almost perfect, whereas the speed is excellent as well. I am writing this using the fluid speech pro model as we speak. So as to say.

    6 likes

  • FluidVoice's profile picture

    @FluidVoiceApp

    Love to hear that :) enjoy!!

    2 likes

  • Brandon Charleson's profile picture

    @brandon_ai

    Simply awesome. Best one out there.

    3 likes

  • FluidVoice's profile picture

    @FluidVoiceApp

    Thank you, Brandon 💙

  • Blueorbit's profile picture

    @blue0orbit

    Intresting

    2 likes

  • FluidVoice's profile picture

    @FluidVoiceApp

    It sure is!!

    1 like

  • tessa's profile picture

    @tessadotsh

    i’d love to try this with low quality radio calls

    2 likes

  • Max Milne's profile picture

    @maxmilneaus

    Well done, round of applause.

    1 like

  • FluidVoice's profile picture

    @FluidVoiceApp

    Thanks, Max :)

  • extracsw's profile picture

    @marwancsw8

    The world is not ready for this

    1 like

  • FluidVoice's profile picture

    @FluidVoiceApp

    Well spread the word and make em aware of it 😉

  • Gabriel Moncha's profile picture

    @gabimoncha

    wow great release. looking to try out the CoreML once it's released

    1 like

  • FluidVoice's profile picture

    @FluidVoiceApp

    It's all released as well :)

  • mike s.'s profile picture

    @twitatwitin

    Please create an iOS app

    1 like