Ming 3.0 TTS Tiny

Native model and prompt implementations adapt inclusionAI/Ming-3.0-tts, copyright (c) 2025 inclusionAI, under MIT.
Kimi Delta Attention follows FLA 0.4.2 equations and numerical conventions, copyright (c) 2023-2025 Songlin Yang, with source credits to Yu Zhang and the Moonshot AI Team, under MIT. No upstream Triton/CUDA kernels are included.
The CAMP+ architecture is from 3D-Speaker, copyright 3D-Speaker, under Apache License 2.0; weights come from the Ming Tiny checkpoint.
Reference resampling follows TorchAudio's Hann-sinc implementation, copyright (c) 2017 Facebook Inc. (Soumith Chintala), under BSD 2-Clause. Log-mel features use Kaldi under Apache License 2.0.
Complete corresponding license texts are retained in CODE-LICENSE.