I have read different research papers on VoIP and algorithms related to VoIP (mostly digital audio signal processing) such as VAD, Sound mixing, etc. I am curious that most of these algorithms take 10ms of data into account. While recording PCM-16bit data from a smartphone (Android OS), we get 40ms of minimum data. So, what is the significance of using 10ms of data?
Hi Khubaib Ahmad ,
It may be related to the fact that the operating system process schedulers works with 10 milliseconds quanta.
Hope this helps!
Hi,
In a few words, the size of the time window is a matter of compromise between resolution in the time and frequency domains. The size of the time window is limited by the non-stationary nature of the sound signal approximated by a piecewise-stationary process. The stationary interval, taking into account the nature of sound extraction, in practice is usually limited to 40 ms.
Best regards
Michael Gordon Thank you very much, sir, could you please explain it in a more basic way?
Dear Khubaib Ahmad,
what is needed to be explained?
Pls. let me know and I could try.
Best regards
P.S. Pls, take a look at this excellent article: https://www.iis.fraunhofer.de/content/dam/iis/de/doc/ame/conference/AES-116-Convention_guideline-to-audio-codec-delay_AES116.pdf
Michael Gordon Sir, Thank you very much for the article. I'll study it. What I want to understand is:
Non-stationary nature of the sound signal approximated by a piecewise-stationary process.
What do you mean by piecewise- stationary process?
Dear Khubaib Ahmad,
sorry for the delay.
Using for information transfer of changes in the duration, attack, attenuation and frequency content of the signal makes the speech signal an non-stationary process.
One of the approaches to constructing some equivalent of one-dimensional spectral density of such a process uses a limited rate of speech formation.
The relatively slow rate of change in the shape of the vocal tract allows us to approximate the speech signal with a filter with characteristics that are constant over an interval of the order of tens of milliseconds with an independent excitation source.
Such an approximation, in turn, allows to consider a speech signal as a process
composed of segments of stationary processes.
Best regards.
M.G.
N.B. A rough approximation of the speech by a two-pole linear prediction model with constant characteristics over an interval of 10 ms was proposed by J. Makhoul, J. Wolf in 1973.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers