I used my ipad to record the words "This. Is. A. Test. This. Is. A. Test.", then downloaded it to my computer as a .MOV file and used ffmpeg to extract audio as WAV. I then used the tuneR package to load the WAV file into R.

sndObj <- readWave('myinput.wav')
plot(sndObj@left, type = "l")

It loads fine and the plot looks ok. The problem comes when I segment the words and calculate DTW distances between all pairs of words (using the DTW package in R).

I can't find any evidence that DTW distance is smaller between the correct pairs than between any other pairing (i.e "this->this" is about the same DTW distance as "this->a". Am I missing any important preprocessing?

Below: sample heat map of actual vs expected DTW distances.

sample heat map of actual vs expected DTW distances.


Usually DTW includes mel-cepstrum feature extraction and cepstrum mean normalization. I'm not sure if you have it.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.