TTS ONNX Text-to-Speech Program using Sherpa ONNX

Why

Because high-quality offline text-to-speech options for Korean, Chinese, and Japanese on Linux are still limited.

What It Is

TTS ONNX is a Linux-first text-to-speech runtime based on Sherpa-ONNX. It supports local/offline synthesis, multilingual voice workflows, direct CLI usage, and HTTP service mode.

Sherpa-ONNX is an open-source speech toolkit from the k2-fsa project. It uses ONNX Runtime for local inference and provides building blocks for speech tasks such as text-to-speech, speech recognition, and related multilingual voice workflows across platforms.

Offline by design

Generate speech locally without requiring cloud inference.

Model lifecycle tools

Install, list, and manage KO/ZH/JA/EN model packages.

Two runtime styles

Use one binary for command-line synthesis or service APIs.

Sherpa-ONNX Project URLs

Model URLs (Known Good)

Korean (Mimic3 VITS): vits-mimic3-ko_KO-kss_low.tar.bz2
Chinese (Piper): vits-piper-zh_CN-huayan-medium.tar.bz2
Japanese + Multi-lang (Kokoro Int8): kokoro-int8-multi-lang-v1_0.tar.bz2
English (Kitten): kitten-nano-en-v0_1-fp16.tar.bz2

How To Use It In CLI

Install models

./bin/tts --install-remote-id vits-mimic3-ko_KO-kss_low
./bin/tts --lang zh --install-remote-id vits-piper-zh_CN-huayan-medium
./bin/tts --lang ja --install-remote-id kokoro-int8-multi-lang-v1_0
./bin/tts --install-remote-id kitten-nano-en-v0_1-fp16

Check available voices

./bin/tts --voice-list
./bin/tts --voice-list --lang ko
./bin/tts --voice-list --lang zh
./bin/tts --voice-list --lang ja
./bin/tts --voice-list --lang en

Synthesize speech

./bin/tts --lang ko "안녕하세요. 오프라인 테스트입니다."
./bin/tts --lang zh "这是中文语音合成测试。"
./bin/tts --lang ja "こんにちは。オフラインテストです。"
./bin/tts --lang en "Hello, this is an offline test."

Save to a WAV file

./bin/tts --out ./out.wav "save this audio"

How To Use It As A Service

Start service directly

./bin/tts --service --config ./config/config.sherpa.yaml

Install user service (systemd --user)

bash ./scripts/install-user-unit.sh
bash ./scripts/enable-user-service.sh

Service health check

curl -fsS http://127.0.0.1:18741/v1/health

Speak via API

curl -X POST http://127.0.0.1:18741/v1/speak \
  -H 'content-type: application/json' \
  -d '{"text":"hello world","lang":"en","format":"wav"}' \
  --output out.wav

List installed models via API

curl -fsS http://127.0.0.1:18741/v1/models

Base service URL: http://127.0.0.1:18741/v1
Key endpoints: /health, /models, /models/install, /speak, /stop, /metrics.

License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0). See LICENSE.txt in this repository for the full license text.

Download

Download Releases