SignalZero Voice Server

The Voice Server is a dedicated microservice for the SignalZero AI ecosystem, providing high-quality Speech-to-Text (ASR) and Text-to-Speech (TTS) capabilities optimized for running on CPU environments (like macOS via Docker).

Features

• ASR (Recognition): Powered by OpenAI Whisper medium.en for high accuracy.
• TTS (Synthesis): Powered by Kokoro-ONNX for high-quality, lightweight neural voice output.
• Voice Activity Detection (VAD): Intelligent silence suppression with a tunable RMS noise gate to prevent hallucinations.
• Wake Word: Mandatory "Axiom" wake word detection.
• Remote Control: FastAPI endpoints to toggle the microphone state ("ears") from the Chat UI.
• Audio Routing: Uses PulseAudio over TCP to bridge Docker audio to the host system (e.g., Bose Flex speakers).

Architecture

The service runs a background recording loop that:

1. Filters noise using an RMS threshold.
2. Detects speech using WebRTC VAD.
3. Transcribes valid utterances with Whisper.
4. Identifies the speaker using SpeechBrain.
5. Routes the message to the SignalZero Kernel if the "Axiom" wake word is present.

It also provides a /speak API used by the Kernel to generate and play neural responses.

Installation & Configuration

This service is designed to be run via the SignalZero-Docker composition.

Environment Variables

• KERNEL_URL: URL of the SignalZero LocalNode (default: http://localnode:3001/api).
• REDIS_HOST: Redis host for state locking (default: redis).
• PULSE_SERVER: PulseAudio server address (e.g., tcp:host.docker.internal:4713).

Hardware Optimization

Currently patched to run efficiently on CPU (ARM64/Apple Silicon) by forcing Float32 precision and ONNX execution providers.

License

This project is licensed under the MIT License.