Speech Swift Delivers Voice AI for Apple Silicon

By vramkickedin | | 2 min read
A close up of a woman speaking in to a microphone in digital mesh decoration

Speech Swift is a comprehensive AI speech toolkit designed specifically for Apple Silicon devices. It allows users to run powerful speech models locally, including tools for speech recognition, text-to-speech synthesis, and full voice conversation capabilities.

Developed by Soniqo, this open-source suite addresses the need for privacy-focused AI tools that operate without an internet connection. It requires no cloud API keys, ensuring that sensitive audio data remains entirely on the user's Mac or iOS device.

Model Size: varies by model & VRAM GPU: 6.5GB required

Comprehensive audio processing tools

  • Speech-to-text recognition supporting 52 languages.
  • Text-to-speech synthesis with streaming and voice cloning.
  • Full-duplex speech-to-speech conversation models.
  • Real-time noise suppression and voice activity detection.
  • Built-in HTTP server compatible with OpenAI Realtime API.

This toolkit serves a wide range of users, from mobile app developers to researchers. Developers creating iOS apps can utilize the lightweight Kokoro TTS model, which runs smoothly on the Neural Engine. Agencies transcribing interviews or meetings can use the high-accuracy Qwen3-ASR models offline, avoiding recurring subscription costs for cloud services.

Optimized for Apple hardware

The project is built natively for the Apple ecosystem, requiring macOS 14 or iOS 17 on M-series chips. Users can install the tools via Homebrew or Swift Package Manager, making integration straightforward for existing projects.

The developers emphasize the privacy benefits of local processing. The documentation notes that the system:

'Runs locally on Apple Silicon — no cloud, no API keys, no data leaves your device.'

For those looking to build their own applications, the source code is fully available and includes example apps for dictation and voice assistance.

Speech Swift can be found on their GitHub project page.