Swift version of parakeet-mlx

[BrandonWeng]

Thanks so much for open sourcing this. We were able to get it running in Python very easily, but trying to integrate it into our Swift app was quite cumbersome - we've started some work on converting the logic from this repo with the help of swift-mlx. There might be some problems still but hopefully it's a start for others looking to use Parakeet in Swift!

https://github.com/FluidInference/swift-parakeet-mlx

[senstella]

Oh wow, this is awesome! Thanks for taking the time to port it over to Swift - that's really cool that you got it done in Swift! Really appreciate it, this'll be such a valuable library for anyone wanting to use Parakeet in iOS/iPadOS apps!

[BrandonWeng]

For those looking for a Swift implementation + CoreML solution, we've added parakeet support to FluidAudio:

https://github.com/FluidInference/FluidAudio

On M2 MB Air with 16GB RAM, this is wht we see with 500MB-1GB of RAM usage

   Dataset: librispeech test-clean
   Files processed: 2620
   Average WER: 2.9%
   Median WER: 0.0%
   Average CER: 1.3%
   Median RTFx: 59.7x
   Overall RTFx: 59.8x (19452.5s / 325.1s)

edit: we support v3 and v2, and its a lot faster than ^ now.