Falafel Wyoming STT
Wyoming speech-to-text server for the Falafel voice-command ASR model (sherpa-onnx), for Home Assistant Assist
About This Project
This project provides a specialized speech-to-text server that bridges the Falafel voice-command model—an ASR engine based on sherpa-onnx and the Parakeet TDT architecture—with the Wyoming protocol. By exposing the Falafel model as a network-accessible service, it allows Home Assistant Assist and other compatible clients to perform local voice processing without relying on cloud-based transcription services.
The server is designed for containerized deployment using Docker and includes scripts for model acquisition from Hugging Face. It is optimized for efficient inference on CPU-based hardware, making it suitable for self-hosted voice automation pipelines where low-latency transcription is required for command execution and natural language processing.