jcsilva/docker-kaldi-gstreamer-server

Containerizes a full-duplex ASR system with decoupled master and worker processes communicating over the network, allowing horizontal scaling across multiple worker nodes. Clients connect via WebSocket to perform real-time speech recognition against Kaldi acoustic models, with YAML configuration specifying model paths and decoding parameters. Supports both co-located and distributed deployments, where workers can register with remote masters to handle inference load independently.

291 stars. No commits in the last 6 months.