SparkAudio/Spark-TTS

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

11006 stars · 1156 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Spark-TTS is an LLM-based text-to-speech system built on Qwen2.5 that generates speech directly from single-stream decoupled speech tokens. This repository provides the official PyTorch inference code and pretrained 0.5B model supporting zero-shot voice cloning and bilingual (Chinese/English) synthesis.

Use cases

• clone a voice from a short audio sample without training
• convert text to natural-sounding speech in Chinese and English
• run zero-shot cross-lingual voice synthesis
• generate speech with controllable parameters like speed and pitch
• integrate LLM-based TTS into a production pipeline
• experiment with decoupled speech token models for research

When to choose

• you need zero-shot voice cloning without per-speaker training
• you want efficient TTS without extra flow-matching or acoustic models
• you need bilingual Chinese-English or code-switching synthesis
• you want an Apache-2.0 licensed TTS with a small 0.5B model

When to avoid

• you need many languages beyond Chinese and English
• you require a fully managed cloud TTS API rather than self-hosted inference
• you need real-time streaming synthesis with guaranteed low latency
• you lack a GPU and need fast inference on CPU only

Facets

library · maturity active

tts llm-inference speech-recognition machine-learning speech-processing artificial-intelligence large-language-models python voice-cloning zero-shot-tts bilingual qwen2 pytorch text-to-speech linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="SparkAudio/Spark-TTS")