Local text-to-speech via sherpa-onnx (offline, no cloud)
The sherpa-onnx-tts skill provides local, offline text-to-speech synthesis using the sherpa-onnx runtime. Unlike cloud-based TTS services, this skill runs entirely on your machine without requiring internet connectivity or sending data to external servers. It leverages the Piper voice models and sherpa-onnx engine to convert text into natural-sounding speech audio files, giving you complete control over your data and ensuring privacy.
This skill supports multiple operating systems including macOS, Linux, and Windows. It uses the ONNX runtime for efficient inference and comes pre-configured to download the Piper en_US lessac voice model, though users can select alternative models from the sherpa-onnx repository. The skill handles the entire setup process including runtime and model downloads, then provides a simple command-line wrapper for generating speech from text. Output is saved as WAV audio files that can be played or integrated into other applications.
This skill is ideal for developers building voice-enabled applications, content creators needing voiceovers, accessibility tools, automation scripts, or any scenario requiring text-to-speech without cloud dependencies. It's particularly valuable for privacy-sensitive applications, offline environments, or projects where low latency and no API costs are important. The offline nature makes it suitable for embedded systems, local assistants, and applications that need to function without internet access.
The sherpa-onnx-tts skill supports macOS (darwin), Linux (x64), and Windows (x64). The installation process automatically downloads the appropriate runtime binary for your platform.
Yes, you can use any compatible model from the sherpa-onnx tts-models releases. Download your preferred model and update the SHERPA_ONNX_MODEL_DIR environment variable to point to the new model directory. If the model directory contains multiple .onnx files, you can specify which one to use with the --model-file flag or SHERPA_ONNX_MODEL_FILE environment variable.
No, sherpa-onnx-tts runs completely offline after the initial setup. Once the runtime and models are downloaded, no internet connection is needed to synthesize speech. This makes it ideal for privacy-sensitive applications and offline environments.
By default, files are stored in ~/.openclaw/tools/sherpa-onnx-tts/ with the runtime in the 'runtime' subdirectory and models in the 'models' subdirectory. You can customize this location by setting the OPENCLAW_STATE_DIR environment variable before installation.
Use the command-line wrapper: sherpa-onnx-tts -o output.wav "Your text here". The -o flag specifies the output WAV file path. You can either run the wrapper directly from the skill's bin directory or add it to your PATH for easier access.
Quick Setup:
.claude/skills/Repository
steipete/clawdis