Tell me more ×
Signal Processing Stack Exchange is a question and answer site for practitioners of the art and science of signal, image and video processing. It's 100% free, no registration required.
up vote 0 down vote favorite

I'm trying to strip an Audio signal to identify different Phonemes in the sample.

I have read a tutorial online that states that the Total Energy at different time intervals can be used to measure and then initialising a threshold can determine if this is White-Noise or if this is a Phoneme, I have used this method.

I have to implement another method for this, and someone has suggested that I count the total number of zero-crossings and then have a threshold value, I agree with this, however, I would like to implement another method.

I was thinking about creating an FFT Algorithm that puts all of the samples into the frequency of which I need and then using a threshold determine which is a Phoneme (i.e. the time interval with the higher frequencies, is considered the Phoneme).

What are the advantages and disadvantages to using an FFT over Zero-crossings?

Thanks :)

share|improve this question
1  
Not clear exactly what you're proposing. A zero-crossing detector would typically be used to estimate the frequency of some periodic signal. A DFT is used to decompose a signal into its constituent frequency components. If you just want to look at the energy content of some frequency band over time, then applying bandpass filter around the frequency of interest would accomplish the same goal. – Jason R Nov 9 '12 at 13:50
@JasonR Thanks for the reply. I have a sample of someone saying "Yes" and I need to pick the different Phonemes out "Y", "E", "S" so I need to take the sample, and remove leading and trailing blocks that do not contain sufficient energy or frequency to be considered. Zero-crossing/energy is good for this, I just wanted a different technique. Any ideas? :) – user1582478 Nov 9 '12 at 13:52
1  
Why do you want a different method? DFTs aren't really well-suited to that sort of algorithm, since they are inherently block-oriented. – Jason R Nov 9 '12 at 13:55
@Jason R So do you think I should stick with Zero-crossings and energy? Will this remove background noise though?? That is my intention, I need to filter out the sample to remove / ignore any background nosie! – user1582478 Nov 9 '12 at 13:59
I suppose you could just look at the spectrogram or STFT of the signal and determine when energy exists, and when it doesnt. Can you throw a plot up of your time domain signal? – Mohammad Nov 9 '12 at 16:12

Know someone who can answer? Share a link to this question via email, Google+, Twitter, or Facebook.

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.