話者分離のSwiftライブラリ。文字起こしも。話者識別はそのままだとできなそうなので改造は必要そう。 | FluidInference/FluidAudio: Native Swift and CoreML SDK for local speaker diarization, VAD, and speech-to-text for real-time workloads. Works on iOS and macOS. github.com/FluidInference…

The CAM++ CoreML conversion I did myself, but the segmentation-3.0 model on CoreML is made possible thanks to the amazing FluidAudio project by FluidInference: github.com/FluidInference…

a few things i have learned about custom vocab support for STT: - the idea sounds promising and useful however custom vocab support is quite limited in usage you can't just create a universal custom vocab list. it needs to be tailored to specific fields - this is all relatively

we are about 50 stars from 1000 , we are so to 4 digits github.com/FluidInference…

we are now 30 stars to 1000 github.com/FluidInference… #SpeechToText #SpeakerDiarization #SwiftSDK #OnDeviceAI

First Reddit Promo Post Let's See where this goes, 20 stars to 1000. reddit.com/r/iOSProgrammi…

WE HAVE REACHED OVER 1,00 STARS on github.com/FluidInference… Special Shoutouts to mannequin Hamza Qayyum Benjamin Lee Sachin Desai & everyone else for their contributions

Thank goodness I live at home without paying rent. These coding agent max/pros subscriptions are killing my wallet

opus 4.5 marked the jump from sde1 to sde2. i’m having a lot of fun and getting a lot of shit done but i’m also realizing we’re going to be the last generation of software engineers

I think Gemini 3 is the best so far, for ML applications it’s able to think the problem through hours and very long context windows without any issues. Even on parts where it failed multiple times. Google might have outplayed themselves with their TPU. Getting the models to be

the most ambiguous part about understandings model is wether you can explain things conceptually or if it was just by happen stance. parakeet eou has 17 encoder layers not out of any conceptual reasons just simply due to test results suggesting 17 is the sweet spot.

thanks to my oss work, it becomes alot easier to accumulate tweeter followers

currently running on M2 2022 mac, gemini3 is too damn slow and eats. up so much of my ram. it doesn't know when to quit which is a flaw cause it would spend hours on something to get nothing working. i think opus 4.5 might be better it gaslights less now.

use coding agents is alot like being a consultant. you don't do alot of actual work but the project can only go as far as you know.