I'm trying to analyse the similarities between the two different audio signals I had used cross-correlation, but I'm finding difficulty in analyzing the correlated output, In general, what are parameters have to be considered for analyzing the audio signals, Is wavelet transform a better option or STFT?
Hello,
To do so, you can use :
1- the cepstral distance
2 - the squared distance of the error computed between the AR coefficients of the two signals
3- the Itakura-Saïto measure that can be computed (a) directely from the PSD periodogram estimates or (b) from the AR (autoregressive) coefficients
4 - the log likelihood ratio distorsion measure
5- a perceptual measure such that the standardized PEAQ (Perceptual Evaluation of Audio Quality)
cross correlation measures similarity between two signals.
For simple audio signals we can use different sampling techniques. Frequency has to be taken in to consideration i.e.ranges. For best we can use laplace transform for analyzing the similarity(synthesis). PSD can also be used.
https://asmp-eurasipjournals.springeropen.com › ...
Autocorrelation-based noise subtraction method with smoothing ...
the following is article by Céline De Looze and Stéphane Rauzy who have used such a technique to define similarity-dissimilarity and convergence-divergence in some prosodic features of dialogue.
https://www.researchgate.net/deref/https%3A%2F%2Fhal.archives-ouvertes.fr%2Fhal-01705534%2Fdocument
It depend by the type of audio signal if the signal is non-linear in time it's better to use windowed wawelet analysis. Easyiest case is if the signal is linear versus time in such a signals you can use both wavelet and FFT, STFT!! Another important issue in order to apply audio signal for wavelet analysis is frequency according Nyquist–Shannon sampling theorem signal should be at least 1 kHz.
I suggest you apply the continuous wavelet transform on each audio signal as they are non-stationary signals. You will get two signals in the time-frequency domain, Following that, find the wavelet coherence between the two scalograms ... then you would notice the high coherence areas in the coherogram that shows high correlation (similarities) between the two audio signal at specified time and frequency (red colour areas), whereas low correlation spots show less similarity between the two signals at a specified time and frequency (blue colour region).
Principly yes, but I have worked on investigation of vasomotion in microcirculation basicaly signal that I investigate is similar like a sound (non-linear), all calculations that I applied, they are done firstable for sounds and according my experience crosscorelations are the best for such a case.
- Badges
- Science topic
- Similar topics
- Indexes