PSIL (Process Speech Impediments Live) is a modular program for speech impediment detection designed to be run as a background process.
Configurations, PortAudio, SampledSignals, ArgParse, Gtk
FFTW, Statistics, LinearAlgebra, Alert
julia psil.jl
The program will print instructions to the command line.
Alternatively, the program can be run with a small GTK-based GUI:
julia psil.jl --gui
This GUI is best ran with at least two threads:
julia -t2 psil.jl --gui
It is, however, recommended to stick to the CLI.
To rerun the configuration process:
julia psil.jl --reconfig
Available modes are:
- (lateral) lisp detection based on Munnich, P., Dederichs, S., Klein, P. & Bruno Figura, C. (2023). Speech Intelligibility: Lateral Lisp Detection for Real-Time Monitoring. Current Directions in Biomedical Engineering, 9(1), 283-286. https://doi.org/10.1515/cdbme-2023-1071:
lisp - basic speech detection via noise gate:
example
The available modes can always be listed using the program:
julia psil.jl --list-modes
Modes can easily be added by creating a directory in the modes folder with the mode's name.
Within this folder, a calibration file, calibrate.jl and an analysis file, analyze.jl should be present.
The calibration algorithm, which should be wrapped inside a function called calibrate, needs to take no import and return the sampling frequency and the arguments to be fed to the analysis algorithm, which should be in an array.
The analysis algorithm, which should be wrapped inside a function called analyze, needs to take an array containing the audio recording, an integer for the sampling frequency, and then the aforementioned arguments returned by the calibration function.
This function needs to return a number. For all recordings within the total recording length, the number returned by analyze are summed up. If the result is greater zero, the notification is sent.
Within the analysis file, a function called default_segment_length needs to exist. As its name implies, this returns the length of audio recording segments that the algorithm function will be run over.
Additionally, a function called analysis_values must exist and return total number of segments to be recorded before the greater zero condition is checked, along with the notification message to be sent out.
An example mode can be found in modes/example.