I have a recording of speech in which I can hear some echo (in a specific vowel). Is there a way to isolate this echo from the acoustic signal and estimate its spectral properties (in Praat for example)?
If you only have a single recording (mixed original speech + echo), this will be difficult without introducing distorsions. But this could be done when analazing the correlation function during the specifix speech segment (vowel).
Another possibility : il you have another file containing the original speech alone, you could try to implement some kind of acoustic echo canceller in order to isolate this echo from the recording of speech.
I hope this will help.
Pascal
Thank you, Pascal. I have a single recording, the echo is part of the original sound (probably due to recording conditions).
I agree with Pascal. In general, I would not even try doing this unless the sample is somehow very valuable. Also, if you can hear the echo in one portion of the sample, it probably is present in the whole signal, unless it is somehow very frequency specific.
Another way to clean the signal - if it actually is worth the trouble - would be to measure the echo properties of the recording space and then try and implement a filter to cance the echo from the recording. But like said, likely to be more work then it's really worth.
Thanks. I was actually interested in characterizing the echo, not necessarily remove it.
You can study your recording in the time frequency plan with, for exemple Wigner transform or fractional Fourier transform. With the fractional Fourier transform it may be possible to isolate the echo from the signal
I'm not an expert at this, but it occurs to me that some cepstral analysis may provide insight, as you have indicated that the echo occurs during a vowel. You may be able to coax some information about the echo out of some longer delays in the reflection coefficient space.
- Badges
- Science topic
- Similar topics
- Correction