Spark-TTS

Free plan LinuxSelf-hostedWebWindows

What Spark-TTS is

Spark-TTS is a free, Apache-2.0 text-to-speech system built around an LLM for voice synthesis. It supports Chinese and English, including cross-lingual and code-switching voice cloning. Speech generation can be adjusted by speaker gender, pitch, and speaking rate. The system uses Qwen2.5 and reconstructs audio from codes predicted by the language model. A locally launched web interface lets users create voices or clone them by uploading reference audio or recording it. The project documents Linux installation and links to a Windows guide, with self-hosted and web use listed alongside Linux and Windows platforms. It also provides a reference for deployment with Nvidia Triton and TensorRT-LLM. The free inference code requires a model download. Intended uses include academic research, education, personalized speech synthesis, assistive technology, and linguistic research. Commercial use is marked unavailable, and the developers warn against unauthorized cloning, impersonation, fraud, scams, deepfakes, and illegal use. Training code and the VoxBox dataset remain listed as pending.

Who it suits

It suits research, education, assistive, and personal speech-synthesis projects that can run self-hosted inference and follow its usage restrictions.

What it does well

Free Apache-2.0 inference code
Supports Chinese and English
Offers cross-lingual voice cloning
Controls pitch and speaking rate
Provides a local web interface

What to know first

Commercial use is unavailable
Model download is required
Training code is still pending
Unauthorized cloning is prohibited

Plans

Plan	Billed and limits	Price
Spark-TTS (Apache-2.0 open-source inference code)	Inference code; model download required	Free

As sparkaudio.github.io lists them, read 12 Oct 2026. The maker's own page has the final word.

Verdict

Spark-TTS provides free self-hosted inference with bilingual synthesis and voice controls. It is aimed at non-commercial uses and requires careful attention to its cloning restrictions and pending training materials.

Facts from Spark-TTS's own pages

Commercial use: No · sparkaudio.github.io, 20 Sept 2026
Deployment: The repository provides a reference for deploying Spark-TTS with Nvidia Triton and TensorRT-LLM. · github.com, 12 Oct 2026
Free plan: Yes · sparkaudio.github.io, 20 Sept 2026
Installation: The project documents Linux installation and links to a Windows installation guide. · github.com, 12 Oct 2026
Intended use: The project describes academic research, education, personalized speech synthesis, assistive technologies, and linguistic research as intended uses. · github.com, 12 Oct 2026
Languages: It supports Chinese and English, including cross-lingual and code-switching voice cloning. · github.com, 12 Oct 2026
License: The repository identifies its license as Apache-2.0. · github.com, 12 Oct 2026
Model design: The system uses Qwen2.5 and reconstructs audio directly from codes predicted by the LLM. · github.com, 12 Oct 2026
Platforms: web,windows,linux · sparkaudio.github.io, 20 Sept 2026
Purpose: Spark-TTS is an LLM-based text-to-speech system for voice synthesis. · sparkaudio.github.io, 12 Oct 2026
Training materials: The repository's to-do list names release of the training code and VoxBox training dataset as pending items. · github.com, 12 Oct 2026
Usage restrictions: The developers warn against unauthorized voice cloning, impersonation, fraud, scams, deepfakes, and illegal use. · github.com, 12 Oct 2026
Voice cloning: It supports zero-shot voice cloning without training data specific to the cloned voice. · github.com, 12 Oct 2026
Voice controls: Speech generation can be controlled through speaker gender, pitch, and speaking rate. · github.com, 12 Oct 2026
Web interface: The locally launched Web UI supports voice cloning by uploading reference audio or recording it, as well as voice creation. · github.com, 12 Oct 2026

Where Spark-TTS stands on Freedom251

Text-to-speech tools#57 of 232
Text-to-Speech Software#73 of 223