IndexTTS
Local text-to-speech software that clones a voice from one recording and supports multilingual speech with emotion and pronunciation controls.
Alternative to:
IndexTTS is a local text-to-speech system for people creating spoken audio and developers adding speech generation to applications. It can reproduce a voice from a single reference recording.
- Multilingual voice cloning: Generate speech in Chinese, English, Japanese, Spanish or Arabic, including a target language different from the reference recording.
- Emotion control: Shape delivery with an emotional audio sample, a written description, the script itself or explicit emotion intensities, separately from the voice reference.
- Speed and pronunciation: Adjust speaking speed and specify pronunciations with Chinese Pinyin, English CMU phonemes or Japanese Kana.
- Application integration: Use the local browser interface for speech generation, integrate through the Python API, or serve the model with vLLM.
The software can use NVIDIA CUDA acceleration. Lower-precision inference reduces GPU memory use. Models are available through HuggingFace or ModelScope, and the first run also downloads some smaller models and example audio.
