lowfilter

i am making a real time prosody modifier using real time TD-PSOLA (Time Domain Pitch Synchronous OverLap Add). My PSOLA is implemented in mxj (unfortunately, i wont be able to share the code because of privacy politics of my lab...)

This code modifies pitch and time of pre-recorded speech signals, using a circular buffer.

Everything works fine, except that, sometimes, very strange artefacts happen. An example of such artefacts can be found there: 

https://soundcloud.com/lowfilter/artefact-example

(it happens at the end where the amplitude decreases)

these artefacts happen at regular time intervals when one period of the original signal is repeated indefinitly (as done in the example). However, this time interval changes depending on the repeated period (I have not sorted out if it depends on the frequency of the original signal yet, will do it soon and let you know).

I imagine it will be very difficult to help without the source code, but maybe someone already had this kind of problem, and have a solution to solve it?

PS: if you are interested in PSOLA, you can find a few articles there:

http://www.sciencedirect.com/science/article/pii/016763939090021Z

http://www.aes.org/e-lib/online/browse.cfm?elib=15382

i am making a real time prosody modifier using real time TD-PSOLA (Time Domain Pitch Synchronous OverLap Add). My PSOLA is implemented in mxj (unfortunately, i wont be able to share the code because of privacy politics of my lab...)
This code modifies pitch and time of pre-recorded speech signals, using a circular buffer.

Everything works fine, except that, sometimes, very strange artefacts happen. An example of such artefacts can be found there: https://soundcloud.com/lowfilter/artefact-example
 (it happens at the end where the amplitude decreases)

these artefacts happen at regular time intervals when one period of the original signal is repeated indefinitly (as done in the example). However, this time interval changes depending on the repeated period (I have not sorted out if it depends on the frequency of the original signal yet, will do it soon and let you know). 

TD-PSOLA: http://www.sciencedirect.com/science/article/pii/016763939090021Z
Real time TD-PSOLA: http://www.aes.org/e-lib/online/browse.cfm?elib=15382


strange-artefact-problem-with-overlap-add-synthesis

So, when I change the pitch of any signal to 440Hz, the time interval between 2 artefacts is about 8 seconds. If the pitch is choosen to be 220Hz, it will take about 4 seconds, and about 16 seconds for 880Hz.

looks like a linear relationship between target frequency and time interval...

So, when I change the pitch of any signal to 440Hz, the time interval between 2 artefacts is about 8 seconds. If the pitch is choosen to be 220Hz, it will take about 4 seconds, and about 16 seconds for 880Hz.
looks like a linear relationship between target frequency and time interval...


Strange artefact problem with overlap add synthesis