Izwi

Izwi is a local first audio inference engine for speech, voice and chat workflows. It runs entirely on the user's own machine, so there are no cloud services, no API keys and no data leaving the device. The listing points to its user documentation site. The documented capabilities cover a broad set of audio tasks. Voice mode supports real time voice conversations with AI, and a text chat is included alongside it. Transcription converts audio to text, diarization identifies and separates multiple speakers, and text to speech generates natural sounding speech from text. Voice cloning creates a voice from a short audio sample, voice design builds custom voices from text descriptions, and a voices section manages built in, cloned and designed voices. The documentation also covers a command line reference, model management and an HTTP and WebSocket API reference with stable, preview and realtime surfaces. Izwi targets developers and privacy conscious users who want speech tooling without a hosted service. The stated minimum requirements are 8 GB of RAM and 10 GB of free storage, with 16 GB of RAM and 50 GB recommended. It supports macOS 12 or later, Ubuntu 20.04 or later and Windows 10 or later. Metal acceleration is used on Apple Silicon with macOS 15 or later, with CPU fallback elsewhere, and NVIDIA CUDA support varies by build. Izwi is open source software released under the Apache 2.0 license. No paid plans are published. Voice cloning works from a short user supplied sample, and the vendor does not state a consent requirement for cloned voices.