hansenj@ele.uri.edu (Jesse Hansen) wrote in message news:<21d5c5c7.0410070955.b827dfc@posting.google.com>...
>
> Also see B. Gold & N. Morgan, "Speech and Audio Signal Processing:
> Processing and Perception of Speech and Music"
>
> The type of speech feature extraction advocated by Morgan and
> colleagues at UCal Berkeley / ICSI is called Perceptual Linear
> Prediction (PLP). On page 299 of the text mentioned above, Morgan
> pre-emphasizes the smoothed power spectrum in the process of
> performing feature extraction. (Figure 22.4 "LPC vs. PLP") If you
> need more details, let me know.
>
> - Jesse
>
> hansenj at ele dot uri dot edu
Thanks for the info. I am waiting on the book from our uni, library. I
will post back after reading.
Reply by Jesse Hansen●October 7, 20042004-10-07
jandouh@yahoo.com (Mohamed) wrote in message news:<e599c748.0410062345.45213acc@posting.google.com>...
> It is common practice in speech research to apply a pre-emphasis
> filter before FFT in order to boost high frequency by 6dB/octave. I
> was told that multiplying the spectrum points by its frequency would
> yield the same 6dB/octave effect on the spectrum. The only mention of
> such method I could locate was in (Oppenheim 1970):
> "If S(w) represents the spectral section to be displayed, the high
> frequency emphasis is accomplished by multiplying S(w) by a function
> L(w), which is constant to some frequency and thereafter increases
> linearly with a specified slope"
Also see B. Gold & N. Morgan, "Speech and Audio Signal Processing:
Processing and Perception of Speech and Music"
The type of speech feature extraction advocated by Morgan and
colleagues at UCal Berkeley / ICSI is called Perceptual Linear
Prediction (PLP). On page 299 of the text mentioned above, Morgan
pre-emphasizes the smoothed power spectrum in the process of
performing feature extraction. (Figure 22.4 "LPC vs. PLP") If you
need more details, let me know.
- Jesse
hansenj at ele dot uri dot edu
Reply by Jesse Hansen●October 7, 20042004-10-07
jandouh@yahoo.com (Mohamed) wrote in message news:<e599c748.0410062345.45213acc@posting.google.com>...
> It is common practice in speech research to apply a pre-emphasis
> filter before FFT in order to boost high frequency by 6dB/octave. I
> was told that multiplying the spectrum points by its frequency would
> yield the same 6dB/octave effect on the spectrum. The only mention of
> such method I could locate was in (Oppenheim 1970):
> "If S(w) represents the spectral section to be displayed, the high
> frequency emphasis is accomplished by multiplying S(w) by a function
> L(w), which is constant to some frequency and thereafter increases
> linearly with a specified slope"
Also see B. Gold & N. Morgan, "Speech and Audio Signal Processing:
Processing and Perception of Speech and Music"
The type of speech feature extraction advocated by Morgan and
colleagues at UCal Berkeley / ICSI is called Perceptual Linear
Prediction (PLP). On page 299 of the text mentioned above, Morgan
pre-emphasizes the smoothed power spectrum in the process of
performing feature extraction. (Figure 22.4 "LPC vs. PLP") If you
need more details, let me know.
- Jesse
hansenj at ele dot uri dot edu
Reply by Mohamed●October 7, 20042004-10-07
It is common practice in speech research to apply a pre-emphasis
filter before FFT in order to boost high frequency by 6dB/octave. I
was told that multiplying the spectrum points by its frequency would
yield the same 6dB/octave effect on the spectrum. The only mention of
such method I could locate was in (Oppenheim 1970):
"If S(w) represents the spectral section to be displayed, the high
frequency emphasis is accomplished by multiplying S(w) by a function
L(w), which is constant to some frequency and thereafter increases
linearly with a specified slope"
Now, my understanding of DSP is minimal but I can learn fast if
directed well :) So if you have any information related to this,
please post back...
Any and all help will be appreciated...
@ARTICLE{Oppenheim1970,
author = {A. Oppenheim},
title = {Speech spectrograms using the fast Fourier Transform},
journal = {IEEE Spectrum},
year = {1970},
volume = {7},
number = {8},
pages = {57-62},
month = {Aug.},
}