Speak Turbo - Talk to your Claude 90ms latency!

Give your agent the ability to speak to you real-time. Talk to your Claude! Ultra-fast TTS, text-to-speech, voice synthesis, audio output with ~90ms latency....

This package appears coherent for a local, privacy-focused TTS service. Before installing, consider: (1) install.sh will run pip install and optionally cargo build — these commands will fetch packages from upstream registries (review pocket-tts and other dependencies before running). (2) The installer writes a CLI into ~/.local/bin and creates ~/.speakturbo (daemon PID and optional config). If you run the daemon, it listens only on 127.0.0.1:7125 but you should still ensure you trust the dependencies (pocket-tts) that implement the TTS model. (3) The output-file allowlist is enforced via realpath checks; be cautious of potential TOCTOU race scenarios when writing files if you allow arbitrary directories. (4) If you need stronger assurance, inspect pocket-tts and the included files locally, run the test-suite, or run the daemon in a restricted environment (container or dedicated user) before giving it broader access.

Type: OpenClaw Skill Name: speakturbo-tts Version: 1.0.7 The OpenClaw AgentSkills skill bundle 'speakturbo-tts' is classified as benign. The code and documentation demonstrate strong security practices, notably the daemon binding exclusively to `127.0.0.1` with DNS rebinding protection (in `speakturbo/daemon_streaming.py`) and robust output path validation for the `-o` flag (in `speakturbo-cli/src/main.rs` and `speakturbo/cli.py`). This validation includes canonicalizing paths to prevent symlink and path traversal attacks, and providing clear instructions for users/agents on how to manage allowed directories. There is no evidence of data exfiltration, malicious execution, persistence mechanisms, or prompt injection attempts in `SKILL.md` or `AGENTS.md`. All network communication is confined to localhost, and `subprocess` calls are for controlled daemon management and audio playback.