Rust Voice Assistant: Whisper + Cerebras + Kokoro TTS

A local, voice-driven AI assistant written in Rust. It captures microphone audio, transcribes it locally, queries a remote LLM, and plays back synthesized speech.

Core Architecture

• Speech-to-Text (STT): Local transcription using kalosm (Whisper Small).
• LLM Inference: Remote processing via the Cerebras API (llama-4-scout-17b-16e-instruct).
• Text-to-Speech (TTS): Local synthesis using sherpa_rs (Kokoro TTS).
• Audio I/O: Microphone streaming via kalosm and audio playback/file generation via rodio and hound.

Prerequisites

1. Kokoro TTS Models: You need the kokoro-multi-lang-v1_0 directory in your project root containing model.onnx, voices.bin, tokens.txt, and the respective dictionaries/lexicons.
2. Cerebras API Key: Required for LLM inference.
3. System Dependencies: A working microphone and system audio output.

Voice Commands

The system continuously listens to your microphone and triggers specific pipelines based on spoken prefixes.

1. File Generation Mode

• Triggers: please generate..., please store..., please say...
• Action: Synthesizes the following text into speech and saves it locally as output.wav.

2. AI Conversation Mode

• Triggers: please ai..., colon colon ai..., talk to me...
• Action: Strips the trigger phrase, sends the prompt to Cerebras, chunks the text response, and streams the synthesized audio back through your default speakers using rodio.

Setup & Run

cargo build --release
cargo run --release