Re: evaluation of speech enhancement algorithm. How?

Hi,

Not exactly. Here are somethings you have to consider:

1) You would have to have access to the original speech and noise as
separate signals if you wish to compare the original AI with a second
AI from your predicted speech, otherwise it won't work. Your estimation
noise input for the second AI is the prediction error e= speeh -
predicted. Your speech input to the second AI would still be the
original speech (if your prediction is good, your error should be
smaller than the original noise).

2) You have to calculate the spectrum of your speech signals and noise.
Average them over 20 discrete bands given by the AI help, and feed them
into the AI algorithm. You can assume the system's orthometric response
is 0.

.

Relevant Pages

  • Re: question about compandents NoiseMuter
    ... Play the speech coder, but don't quantize any of the parameters ... if you are familiar with what 'musical' noise ... I have worked on codecs, doing quick listening tests between changes, where I thought things were going OK. ...
    (comp.dsp)
  • Re: question about compandents NoiseMuter
    ... Play the speech coder, but don't quantize any of the parameters ... if you are familiar with what 'musical' noise ... Further listening ...
    (comp.dsp)
  • Re: help needed: signal processing
    ... I need your help to solve 2 problems with my filtering ... > noise from a .wav audio file. ... Processing speech is a challenging problem. ... to consecutive FFT blocks so there are samples in the FFT that change from ...
    (comp.soft-sys.matlab)
  • Re: LPC analysis question
    ... I conducted an experiment in MATLAB related to speech enhancement. ... I did LPC analysis of a frame of X and obtained an estimate of a noisy ... clean excitation signal u_clean. ... Adding noise will be evident as it ...
    (comp.speech.research)
  • Re: Speech detection (Formulation Problem)
    ... impossible to detect speech in noise in quick and reliable way. ... signal into frame of 20 ms using half-overlapping hamming window. ... threshold, it will remain as speech else it will be reclassified as noise. ...
    (comp.dsp)