- Category
- AI Tools
- Rank
- No. 1554Tools index
Previous survey · No. 1566 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- moeru-ai
- GitHub
- 13 stars
- Latest release
- v0.1.0-beta.4
- Date
About
Local text-to-speech inference server powered by ONNX Runtime, supports Chatterbox and other TTS models.
What it does
ORTTS exposes speech synthesis through an OpenAI-compatible HTTP service. Clients send text, a model choice, and optionally reference voice audio, then receive a WAV file. It includes interactive API documentation and recognizes Chatterbox Multilingual and Kokoro models.
Why it's ranked here
The appeal is practical interoperability: existing clients can target a familiar speech endpoint while inference stays under the operator’s control. Docker startup, source builds, reusable model caching, graceful shutdown, and embedded API documentation make it credible for local service workflows. The unfinished direct-run command keeps the scope narrower than the command surface suggests.
What's good
The project offers two distinct speech backends behind one API contract. Model cache tooling reports identifiers, revisions, disk usage, and modification age, while removal targets recognized models. Linux and Windows receive platform-specific acceleration support, and structured API errors follow the same general shape clients expect from OpenAI-compatible services.
Tradeoffs
Only Chatterbox Multilingual and Kokoro are recognized by the model registry. The direct model-running command is explicitly unfinished. macOS CoreML activation remains disabled in the target configuration. Removing a missing model triggers a panic, and successful removal recursively deletes that model’s cached directory, so operators should treat cache management carefully.
How to use it well
Use it when an application already speaks the OpenAI audio API and needs a locally operated speech endpoint. Start with Docker for a persistent service, mount the Hugging Face cache across restarts, and inspect the embedded documentation before integration. It does not yet cover direct one-shot model execution through its command line.
Technical notes+
Cargo.toml defines a Rust 2024 workspace requiring Rust 1.89, marks the package publish = false, and wires Linux to CUDA plus Windows to DirectML through ortts_onnx; the macOS CoreML target block is commented out. src/cli.rs exposes serve, list, remove, and unfinished run commands. src/main.rs dispatches them, while src/commands/serve.rs binds an Axum listener and installs Ctrl+C and Unix termination shutdown handling. crates/server/src/lib.rs mounts Scalar at the root and publishes /openapi.json. src/utils/available_model.rs maps two accepted model names to Hugging Face repositories. src/commands/list.rs reads cache metadata, and src/commands/remove.rs recursively removes matched model directories.
Observed
- License
- MIT
- Primary language
- Rust
- Packaging and install surface
- Runs from source with Cargo or as a container image from GitHub Container Registry; Cargo publishing is disabled.
- Interfaces
- Command-line interface plus an OpenAI-compatible HTTP API with embedded Scalar documentation and an OpenAPI JSON document.
- Model backends
- Chatterbox Multilingual and Kokoro.
- Platform acceleration
- Linux is configured for CUDA and Windows for DirectML; the macOS CoreML target configuration is commented out.
- Repository structure
- Multi-crate Cargo workspace separating server, ONNX, shared code, and model backends.
Read from README.md, Cargo.toml, src/cli.rs, src/main.rs, src/utils/mod.rs, src/commands/mod.rs, src/commands/list.rs, src/commands/serve.rs, src/utils/dir_size.rs, src/commands/remove.rs, src/utils/available_model.rs, crates/onnx/src/lib.rs, crates/server/src/lib.rs, crates/shared/src/lib.rs, crates/shared/src/error.rs.
What it can do
Convert text to speech
Text string → Audio file or stream
Run local TTS inference
Text input and model selection → Synthesized speech audio
Serve TTS requests via API
HTTP requests with text payload → Audio response
Load and execute ONNX TTS models
ONNX model files → Loaded TTS model for inference
Support Chatterbox TTS models
Chatterbox model files and text → Generated speech audio
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.