speech-swift

A full speech stack that runs locally on Apple Silicon using MLX and CoreML: ASR, TTS, speech-to-speech, voice activity detection, diarization, and denoise. No audio leaves the device, which for meeting and call recordings is not a nice-to-have, it is the whole reason to build it.

Pay for what you use

Each model is its own Swift Package Manager library. If you only need transcription you do not pull a TTS model and its weights into your app. You compose the pieces you actually use and nothing else.

There is a CLI too, so you can try it before you write any code:

speech transcribe meeting.wav
speech diarize meeting.wav --engine sortformer --rttm
speech vad meeting.wav