I want to build a speaker independent voice-controlled system. If I ask 10 people to say one unique word(for example"hello"), and record their voice and do LPC(linear predicting coding) upon them, I have 10 arrays of LPC coefficients, right? So how should I combine these 10 arrays to build one array(my model of "hello word" and use it to compare with my system's new input voices? (I like to use DTW algorithm for this comparation).


  • $\begingroup$ Averaging(or any other operation to get cumulative effect of 10 vectors) the speech features (LPC coefficients) does not work. DTW should be calculated with each of the feature vector. $\endgroup$
    – Arpit Jain
    Jan 16, 2018 at 5:38

1 Answer 1


DTW does not really work in speaker-independent systems, you'd better use a proper speech recognition toolkit.

For 10 recordings of a single speaker, you compare to all one by one and select the best match.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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