whereismyai
Macvoice
Macvoice is an offline voice agent loop for Apple Silicon that runs locally in under 2 GB.
Macvoice is a local voice agent loop designed for Apple Silicon Macs that operates offline under a 2 GB physical memory footprint without PyTorch. It uses Silero VAD for voice activity, Smart Turn for end-of-turn detection, Parakeet TDT for speech to text via MLX, and Phonon for text to speech. The system connects to any OpenAI-compatible server or coding agent over ACP as its brain. Built-in echo cancellation uses Apple's voice processing via AVAudioEngine to allow interruptions while the agent speaks. It targets developers and users wanting a local voice interface.
Macvoice at a glance
Key features
- Runs offline locally in under 2 GB
- Uses Silero VAD and Smart Turn
- Transcribes speech with Parakeet TDT via MLX
- Synthesizes speech using Phonon
- Connects to OpenAI-compatible servers or ACP agents
- Provides built-in echo cancellation and barge-in support
- Includes status line monitoring
Use it for
- Running a local voice agent on Apple Silicon
- Interacting with local LLMs using voice
- Testing offline speech-to-text and text-to-speech pipelines
- Connecting voice input to coding agents over ACP
Alternatives to Macvoice
- MLX Audio — It enables fast audio model inference directly on Apple M-series chips.
- FluidAudio — Runs local audio AI models on Apple Neural Engine.
- Audio Chat Box — Runs entirely offline and locally with sherpa-onnx and no data sent anywhere.
- Lavoi — Free on-device transcription that works locally without a word limit.
- Speech-to-Speech — Fully modular, swappable components with local or hosted execution options.
- AssemblyAI — Industry-leading speech models deliver high accuracy and low latency for voice applications.
- Samples — Generates local audio tracks and sound effects with Stable Audio 3.
- Argmax OSS — Runs AI speech models locally on Apple silicon.
- Free ASR — Zero-dependency real-time streaming speech-to-text with no account login required.
- audio.cpp — Runs modern audio models locally with high-performance, native C++ execution.
- whisper.cpp — It provides high-performance, offline speech recognition on diverse hardware.
- Silero Models — Fast, portable, and simple text-to-speech models for CPU and GPU usage.
All Macvoice alternatives