A VST plugin that automatically detects and cuts unwanted tics sound by using SVM.
- Get the newest binary from Releases;
- pick an appropriate version and load it into your VST host, or use the stand-alone executable version.
Train a new model:
- Prepare audio clips in WAV format (
*.wav) and place tics and non-tics audio samples in two separate folders respectively; - Load the plugin and open the GUI, or start the stand-alone version;
- In the GUI, click the "Train new model" button;
- Select the folders containing tics and non-tics samples, and then training will start;
- After finished, use the model right away and/or save the model by click the "save model to file" button.
Use ThreadCutter to process audio:
- Load the plugin and open the GUI, or start the standalone version;
- Train a new model by abovementioned steps or load a model by clicking "load model from file" button;
- Adjust the gate slider and the detection threshold slider (detailed documentation to be added; for now, refer to "how it works" below);
- Feed audio in and it should work.
Saving and loading presets on the current version has not been implemented yet. Only the model can be saved and loaded as of now.
(To be improved...)
- The gate value is compared to the ratio of current peak level and the running average of peak levels;
- Detection is done by using MFCC for feature extraction, and SVM for classifying;
- The detection threshold value is compared to the proportion of positive matches in a frame.
Visual Studio 2017 project files are under Builds/VisualStudio2017/. For other toolchains, project files may be generated by ProJucer using ThreadCutter.jucer.
To build VST2 target, VST2 SDK must be separately obtained first.
External libraries under ExternalCode:
- Add more adjustable parameters to GUI;
- Performance tuning;
- Clean-up code.