MLX Audio Swift
A modular Swift SDK for running text-to-speech, speech-to-text, voice activity detection, and speech-to-speech models on Apple Silicon through MLX, with importable modules for just the models an app needs.
MLX Audio Swift is a Swift SDK for audio processing built on MLX, Apple's array-computing framework, organized as separate modules so an app can import only the parts it needs. Separate modules cover audio codecs, text-to-speech models such as Qwen3-TTS and Orpheus, speech-to-text models such as Whisper and Parakeet, voice-activity detection and speaker diarization, and speech-to-speech models, alongside a base module of shared types and a set of SwiftUI components for building audio interfaces.
It is added to a project through the Swift Package Manager, importing only the product modules a given feature needs, such as text-to-speech or speech-to-text. Usage examples cover generating speech from text, transcribing an audio file, diarizing speakers, and streaming generation events such as tokens, audio chunks, and status updates as they arrive.
It targets Apple platforms and is written in Swift under the MIT license.