GSV-TTS-Lite and GPT-SoVITS
GSV-TTS-Lite is an optimized inference engine for the GPT-SoVITS model that RVC-Boss/GPT-SoVITS implements, making them complementary tools where the former provides efficient deployment of the latter's trained models.
About GSV-TTS-Lite
chinokikiss/GSV-TTS-Lite
GSV-TTS-Lite A high-performance inference engine specifically designed for the GPT-SoVITS text-to-speech model.(few shot voice cloning)
Implements millisecond-level latency through deep optimization techniques including Flash Attention support and decoupled timbre-emotion control, achieving 3-4x speedup on consumer GPUs while halving VRAM requirements. Provides multiple inference modes (streaming token-level output, batch processing, voice conversion) with subtitle timestamp alignment, and ships as a PyPI package supporting CUDA, MPS (Apple Silicon), and CPU backends via Python SDK, REST API, and WebUI interfaces.
About GPT-SoVITS
RVC-Boss/GPT-SoVITS
1 min voice data can also be used to train a good TTS model! (few shot voice cloning)
Combines GPT language modeling with SoVITS vocoding to enable zero-shot TTS from 5-second samples and cross-lingual inference across English, Japanese, Korean, Cantonese, and Chinese. The WebUI integrates voice separation, automatic dataset segmentation, and ASR labeling to streamline training data preparation, achieving real-time inference speeds (RTF 0.028 on RTX 4060Ti) with minimal compute requirements.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work