argmax-oss-swift
On-device speech-to-text, speaker diarization, and text-to-speech for Apple Silicon.
Swift inference frameworks for Whisper, Pyannote, and Qwen-TTS running locally on macOS and iOS.
This is a collection of turn-key Swift frameworks designed for running speech AI models directly on Apple Silicon devices. You should use this if you are building native iOS or macOS applications that require local transcription, text-to-speech, or speaker diarization without relying on cloud APIs.
What it is
The SDK packages three core inference engines into Swift library products: WhisperKit for OpenAI Whisper speech-to-text, SpeakerKit for Pyannote speaker diarization, and TTSKit for Qwen-TTS text-to-speech. It supports memory-efficient streaming for large audio files, custom voices, and local server execution via an optional CLI.
Who it's for
It suits iOS and macOS developers building offline-first audio, video, or transcription applications requiring local processing on Apple hardware. It does not suit teams building non-Apple applications without wrapping the local server, or those needing pre-built Android support without upgrading to the Pro SDK.
Key features
WhisperKit Speech-to-Text
Run OpenAI Whisper models locally on Apple Silicon for multilingual transcription.
Incremental Audio Loading
Stream large audio files from disk in bounded-memory chunks using VAD silence boundaries.
SpeakerKit Diarization
Perform speaker diarization using Pyannote-based models natively in Swift.
TTSKit Text-to-Speech
Generate speech locally using Qwen-TTS with support for custom voices and real-time streaming.
Up and running in 3 steps
Our honest verdict
What's great
- Runs entirely on-device on macOS and iOS with no cloud API dependency
- Memory-efficient streaming handles large audio files without crashing
- Modular architecture lets you import WhisperKit, TTSKit, and SpeakerKit individually
- Includes Swift CLI tools and local server capabilities for testing
Where it falls short
- Requires macOS 14.0+ and Xcode 16.0+ to build and run
- Advanced features like real-time multi-speaker transcription require the paid Pro SDK
- Running large models locally demands significant Apple Silicon memory and compute resources
Pricing & license
Free under its MIT license for the open-source Swift SDK. Additional advanced features, custom vocabulary, and cross-platform support are available through the commercial Argmax Pro SDK.
Open remote AI-training roles
876 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
Use it, or have us set it up
FAQ
What platforms are supported by this SDK?
The open-source Swift SDK supports macOS 14.0 and later, requiring Xcode 16.0 or later.
Can I process large audio files without running out of memory?
Yes, WhisperKit includes an incremental audio loading mode that streams files from disk in bounded-memory chunks split at VAD boundaries.
Are all features included in the open-source release?
The open-source SDK provides core inference frameworks, while advanced features like real-time transcription with speakers and custom vocabulary require the Argmax Pro SDK.
