Articulatory Synthesis with Evolutionary Computing in Python (ASECP) is a tool for performing Acoustic to Articulatory Parameter Inversion (AAPI) with Praat and Genetic Algorithms.
AAPI refers to ascertaining parameters for a physical model of the vocal tract, that match a given target sound.
For an overview of command line arguments navigate to the root of the directory and type
python main.py -h or
python main.py --help
The simplest usage of the program would be to accept the default values and just specify a target sound file
python main.py -sf "vowel-01.wav"
The default generation size and population size can be overridden by using -gs and -ps respectively. Also please note that the indexing of generations starts at 0 and is inclusive. Gen0 contains the randomly initialised population and Gen1 contains the first population created by the genetic operators.
python main.py -gs 10 -ps 5
Elitism
Elitism is not enabled by default but can enabled and the number of top indiviudals to be kept for the next generation with the following commands.
python main.py -el True -es 4
Selection
Linear, exponential, fitness proportional.
python main.py -sl 'linear'
Sampling
To sample the probability distribution three sampling methods are available.
- Stochastic Universal Sampling
- Roullete Wheel Sampling
- Tournament.
Stochastic Universal Sampling is the default sampling method. Currently, if RWS, or Tournament are required the call to the respective function needs to be changed within the crosover functions.
Crossover
A choice between one point crossover and uniform crossover is available.
python main.py -cr 'one_point'
python main.py -cr 'uniform'
Mutation
The mutation rate and standard deviation for the gaussian distribution can be overridden by using -mr and -sd respectively. The mutation rate should be set between 1.0 and 0.0.
python main.py -mr 0.05 -sd 0.1
Can use script_maker.py to batch runs. Run with cmd.exe /k ..\batch_scripts\batch.bat