A self-hosted speech server with streaming transcription, speech synthesis, verified local model setup, WebSockets, HTTP, gRPC and MCP APIs.
View on GitHub ↗Homepage ↗Impossible Voice is a ready-made, self-hosted speech-to-text and text-to-speech server. The first release targets one curated local STT model and one curated local TTS voice with automatic, checksum-verified installation and offline operation after setup.
The public v0.1 promise is frozen in docs/product-contract.md. HTTP, realtime WebSocket, gRPC, and bounded MCP transports are implemented. Native Windows and Linux archives are built by the release workflow; a prebuilt Linux container is not published in v0.1.
From a source checkout, install the pinned local artifacts with one command, then start the server with one command.
setup installs exactly one pinned sherpa-onnx CPU runtime, one English streaming STT model, and one English TTS voice into the ignored runtime-artifacts directory. Downloads are HTTPS-only, size- and SHA-256-verified, safely extracted into staging, fully inventoried, then atomically activated. setup --offline re-verifies an existing installation without network access; status is read-only. Ordinary serve never downloads artifacts.
The release workflow publishes a Windows ZIP and a mode-preserving Linux tar.gz, each with a SHA-256 sidecar. They contain the native server, scripts, configuration, documentation, and notices—but no runtime or model bytes. See docs/operations.md for archive use, checksum verification, offline restart, and recovery.
serve binds HTTP/WebSocket/MCP to 127.0.0.1:8080 and gRPC to 127.0.0.1:50051 by default. Configuration precedence is defaults, then an optional TOML file, then IMPOSSIBLEVOICE environment variables, then expl
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/impossibleG--impossible-voice.html)
A self-hosted token-completion server for local open models, with chat and completion APIs, SSE streaming, WebSockets, gRPC, MCP and verified setup.
impossibleG/impossible-embeddingsA self-hosted embedding server with HTTP, gRPC and MCP APIs, verified model installation, batching, health checks, metrics and Windows and Linux releases.
impossibleG/impossible-ocrA self-hosted OCR server that turns PNG and JPEG images into structured text using local PP-OCRv5 models through HTTP, gRPC and MCP.
mahingbun-dev/plugin-hubRust 插件中台：契约治理 · 注册发现 · 声明式编排 · 事件总线 · MCP 工具面 / Out-of-process plugin hub in Rust: contracts, discovery, DAG orchestration, event bus, MCP tool surface
vericle/heyclaw🦞 Headless real-time Python voice assistant powered by ElevenLabs, Gemini, Mem0, MCP tools, and local wake-word detection.
montezuma-p/harkenYour agent cannot listen to audio. harken fixes that — locally. MCP server + CLI, one 13 MB binary, fully offline: no Python, no ffmpeg, no API key.
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.