I know that one of the solutions is de-essing (to remove sharp S sounds or "wet speech sounds") but it doesn't seem to work in my case.

The waveform is the following for the lisp part:

enter image description here

I noticed that the "normal" speech looks more regular:

enter image description here

Whereas the "lisp" part is a more "fuzzy" sine wave:

enter image description here

I remember something from the Signals and Systems course at University that the "lisp" part can be "combed out" but I am unable to recall the key terms (and it was on another language).

What are the terms or exact literature I should look into? (integrators, low pass filters, Fourier transform?)

edit: The images are from Audacity.

  • I would say focus less on what the waveforms look like, and look up de-essing using EQs or Multi-band compressors. I'll try to come back and make an answer, but I'm just on break right now.
    – user22688
    Oct 17, 2017 at 20:59

1 Answer 1


A comb filter is a specific type of filter which operates by adding a delayed version of the signal back to itself. The frequency response has a repeating pattern of notches or spikes, like a comb, hence the name. The signal difference you show above is sort of like that in that it is mostly low-pass filtering -- the "normal" speech sine wave is of low frequency compared to components of the "lisp part" one, but there is still high frequency content in the "normal" signal.

Generally deessing is implemented by using bandpass filtering to isolate frequencies between ~2kHz to ~10kHz and using that signal to control a compressor such that those frequencies are made less prominent but not constantly attenuated. There are many ways to implement a band pass filter. (See: https://en.wikipedia.org/wiki/De-essing .)

A Fourier transform converts between time domain and frequency domain. Time domain is what you are looking at in your signals above. Frequency domain is another very useful way to visualize a signal. This page describes the spectral display function in Audacity: http://manual.audacityteam.org/man/spectrogram_view.html . (You can get more information on the underlying algorithm for making such a display by looking up the "short-time Fourier transform" or STFT.)

Also, generally "smoothing" in signal processing means some form of low-pass filtering. With images, we call this "blurring." In practice, we want the smoothing to have some useful properties for preserving content -- such as being able to understand the speech after removing noise. The preservation requirements usually require something more sophisticated than straight low-pass.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.