Kokoro-82M
A high-performance multilingual TTS model with only 82M parameters, Apache 2.0 licensed for commercial use, running on just 4GB VRAM.
Kokoro-82M achieves impressively high-quality speech synthesis with only 82M parameters, trained on less than 100 hours of data. This makes Kokoro a benchmark model in the lightweight TTS category — its extremely low parameter count means it can run efficiently on resource-constrained devices while maintaining satisfactory synthesis quality.
In terms of performance, Kokoro-82M achieves a 32x real-time factor on NVIDIA T4 GPUs, and can still reach 2–3x real-time factor on CPU. Memory usage is approximately 400MB, and it runs smoothly with just 4GB of VRAM, making it an ideal choice for edge deployment and high-concurrency services. The model supports multiple languages including Chinese, English, Japanese, Korean, and French.
Kokoro-82M is released under the Apache 2.0 license and may be used free of charge for commercial purposes. Whether deployed on IoT devices or as a backend engine for large-scale speech synthesis services, Kokoro delivers stable voice output at minimal cost. Visit the Kokoro GitHub Repository.