(Unfinished README)
===
This project is an attempt at making a software capable of analysing speech signals in real-time. The main focus as of writing this description is on pitch detection and formant analysis. Feel free to report any bugs, request new features, or critique existing ones in the Issues section.
You can grab the latest release for your operating system in the Releases page.
If you are on Windows, the release should ship with its dependencies.
If you are on GNU/Linux, jump to the Dependencies section.
After unzipping the release, run the speech_analysis executable.
The graph represents frequencies (vertical axis) in function of time (horizontal axis).
Audio is analysed in a succession of short frames (15-60 ms). One frame equals one "unit" on the horizontal axis.
Formant analysis is performed on every frame. Moving the cursor at a particular point in time lets you read the formant frequencies in more detail.
Pitch detection has the added benefit of functioning as a voice activity detector.
Unvoiced frames will display formant tracks in dim red.
Voiced frames will display pitch as a light blue marker, and mark formant tracks with color-coded circles.
Pressing P lets you pause the analysis.
Pressing S lets you toggle between linear scale and Mel scale.
Holding (Has no effect as of now)R lets you view the raw formant data.
Holding L and pressing Up or Down lets you adjust the order of the linear predictive coding analysis.
Holding F and pressing Up or Down lets you adjust the maximum frequency of any expected formant.
SDL2
SDL2_gfx
SDL2_ttf
PortAudio19
Eigen-unstable
FFTW3
LibTorch