v0.2 for Windows 10/11 64bit

Features

• Completely rewrote the preprocessing pipeline in C++, eliminating dependencies on librosa, numpy, and matplotlib for spectrogram generation.
• Trained using the full frequency spectrum, rather than the previous 2400-20000Hz range.
• Added identification capabilities for DSEE / DSEE HX post-processing modes, covering all operational modes of DSEE HX.
• Uses model capabilities as a prior for hypothesis testing instead of a simple hard threshold, reducing the misjudgment rate.
• All dependencies are statically linked (including ONNXRuntime), resulting in a single portable executable file.
• Compiled with VC-LTL for better compatibility.

Notes

• The model is incompatible with the Python codebase (the change in training frameworks resulted in a different preprocessing pipeline; however, it can be loaded and used via HF/Hugging Face ecosystem tools).
• Old models are incompatible with the new CLI version (for the same reason as above).
• This is a shared program, and there are currently no plans to open-source it.

Legal Disclaimer

• "AS IS" Provision: This software is provided "AS IS", without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose, and noninfringement.
• Unknown Unknowns: Please be explicitly aware that this program may contain unforeseeable and unknown issues ("unknown unknowns"). In particular, the DSEE / DSEE HX detection feature is highly experimental and may exhibit unpredictable behavior, inaccuracies, or unexpected results under certain conditions.
• Limitation of Liability: In no event shall the author(s) or copyright holder(s) be liable for any claim, damages, or other liability, whether in an action of contract, tort, or otherwise, arising from, out of, or in connection with the software or the use or other dealings in the software.

model file

v0.1

Update README.md