Sherpa-ONNX TTS
Cross-platform TTS using ONNX Runtime for on-device speech synthesis.
About
Text-to-speech is one part of sherpa-onnx, an on-device speech toolkit from the k2-fsa next-gen Kaldi project that runs entirely offline through ONNX Runtime. On the synthesis side it executes models from several engines, including VITS, Piper, Matcha, Kokoro, and MeloTTS, producing multilingual speech without any cloud service; the same framework also covers speech recognition, voice activity detection, speaker identification and diarization, keyword spotting, and speech enhancement. Portability is the project's defining trait: it runs on x86, 32-bit and 64-bit ARM, and RISC-V processors, on Linux, Windows, macOS, Android, iOS, HarmonyOS, and WebAssembly, and on boards like Raspberry Pi and NVIDIA Jetson, with NPU support for several vendors. Bindings exist for C++, C, Python, Go, C#, Java, Kotlin, JavaScript, Swift, Rust, Dart, and Object Pascal. Open source under Apache 2.0, it is a common choice for embedding offline speech synthesis in mobile apps, desktop software, and embedded devices where privacy or connectivity rules out hosted APIs.
Should you use Sherpa-ONNX TTS?
Pick it when
Pick Sherpa-ONNX when you must ship offline speech synthesis inside an Android, iOS, HarmonyOS, desktop, or WebAssembly app, or on Raspberry Pi and Jetson boards, and want the same toolkit to cover recognition and VAD too.
Look elsewhere when
Skip it if you want large LLM-based voices such as Orpheus or CosyVoice, which run on their own Python stacks rather than its ONNX lineup of VITS, Piper, Kokoro, Matcha, and MeloTTS. For one model on a Python server, call Kokoro directly.
Alternatives to Sherpa-ONNX TTS
- Piper TTS
Single-purpose offline TTS with many ready voices and a simple CLI, less portable across mobile platforms but easier on a Linux box.
- Kokoro TTS
Using its Python library directly gives streaming and simple setup on CPU or GPU when you do not need mobile bindings.
- Supertonic
Also ONNX on-device with SDKs for many languages and 31 spoken languages, but a single model whose developer has stopped official support.
- eSpeak NG
Tiny formant engine for 100+ languages with no model files, robotic sounding but minimal where even ONNX models are too heavy.
Reviews (0)
Leave a Review
No reviews yet. Be the first to review!
Details
- Category
- Text-to-Speech (TTS)
- Price
- Free
- Platform
- Local/Desktop
- Difficulty
- Easy (2/5)
- License
- Apache-2.0
- Added
- Apr 3, 2026