I am very surprised/ pleased by the quality of the LPC data used in this project. They sound much better than mine (also processed by your python_wizard) (I also noticed that 'ready', 'fail' and 'retard' sounds different)
Can you share the script to generate the sound wave as you did for make_sample.sh? Which voice are you using? (Tom ?)
I also use say on Mac and prepare the sound with Sox.
I have a Python script to process the text file line by line and generate an LPC file that my Arduino can read directly from a sd card
My flow goes like this for each line
# generate temporary audio wave with the Mac's voice
system('say -v"tom" -o audio/' + label +'a.wave -r 140 '+ '[[slnc 10]]'+line)
# convert wave to 8kHz 16 bit wav
system('sox audio/'+ label+'a.wave -r 8k -b16 audio/'+label+'.wav ')
#play audio
system('afplay audio/'+ label+'.wav')
#print label to file so the Arduino can file the line in the file
system('printf '+label+': >> '+ out)
#process wave with python_wizard
system('python python_wizard.py -u 0.93 -w 2 -S -p -a -0.9375 -m 0.9 -F 25 -f hex audio/'+ label+'.wav >> '+ out)
#delete temporary wave
system('rm audio/'+label+'a.wave')
Maybe the quality produced on my side is related on how I use python_wizard but I want to make sure that I generate the wave file correctly before processing them with python_wizard