I'm trying to do frequency detection on a class of devices whose PCM
encoder applies a high pass filter. My frequency detection algorithm works well
unless I try and do frequency detection for the lower notes that are dampened by
the HPF. 

Determining the frequency with the highest peak should work fine for the types
of sounds I use (the fundamental frequency is always the loudest in my
environment). However, in the case where the fundamental frequency lies within
the dampened region of the HPF, its amplitude shows up on the spectrum as being
much lower than its harmonic frequencies. 

So I tried applying harmonic product spectrum analysis. While there was some
improvement over highest-peak detection, the harmonic frequencies are still
being detected for most sounds with a fundamental frequency in the dampened
region. With HPS, it seems to just detect the first major harmonic outside of
the dampened region.

I've eliminated the possibility that the error in detection is due to aliasing.


The HPF seems to lower the amplitudes on a sliding scale by up to 50dB (the
lower the note, the greater the reduction). I've attempted to reverse the HPF by
applying a low pass filter appropriately, but it's too difficult to determine
the amount of amplitude reduction occurring. The HPF used on each device is
slightly different, meaning I need to make a unique LPF for each device, which
is just not feasible.

I would have given up trying to detect the frequency of the lower notes on these
devices, however I see other applications are able to detect lower frequencies
flawlessly. My question is, what frequency detection algorithm could these other
applications using? Any advice?

Thank you,
Scott

______________________________
New Code Sharing Section now Live on DSPRelated.com. Learn about the Reward Program for Contributors here.



(You need to be a member of audiodsp -- send a blank email to audiodsp-subscribe@yahoogroups.com )

Scott-

The normal procedure would be frequency compensation; i.e. measure the transfer
function of the device and apply an
inverse filter prior to specific processing.  For the detection you describe,
this should work assuming you can
tolerate the reduced SNR in the area of the HPF.

But you say that your device has a non-deterministic HPF that varies per device.
 This would have to be substantial
variation for you to be unable to compensate with a fixed inverse filter.  What
kind of device have you got there?

-Jeff

> I'm trying to do frequency detection on a class of devices whose PCM
encoder applies a high pass filter. My frequency
> detection algorithm works well unless I try and do frequency detection for
the lower notes that are dampened by the
> HPF.
>
> Determining the frequency with the highest peak should work fine for the
types of sounds I use (the fundamental
> frequency is always the loudest in my environment). However, in the case
where the fundamental frequency lies within
> the dampened region of the HPF, its amplitude shows up on the spectrum as
being much lower than its harmonic
> frequencies.
>
> So I tried applying harmonic product spectrum analysis. While there was
some improvement over highest-peak detection,
> the harmonic frequencies are still being detected for most sounds with a
fundamental frequency in the dampened region.
> With HPS, it seems to just detect the first major harmonic outside of the
dampened region.
>
> I've eliminated the possibility that the error in detection is due to
aliasing.
>
> The HPF seems to lower the amplitudes on a sliding scale by up to 50dB (the
lower the note, the greater the
> reduction). I've attempted to reverse the HPF by applying a low pass filter
appropriately, but it's too difficult to
> determine the amount of amplitude reduction occurring. The HPF used on each
device is slightly different, meaning I
> need to make a unique LPF for each device, which is just not feasible.
>
> I would have given up trying to detect the frequency of the lower notes on
these devices, however I see other
> applications are able to detect lower frequencies flawlessly. My question
is, what frequency detection algorithm could
> these other applications using? Any advice?
>
> Thank you,
> Scott

______________________________
New Code Sharing Section now Live on DSPRelated.com. Learn about the Reward Program for Contributors here.



(You need to be a member of audiodsp -- send a blank email to audiodsp-subscribe@yahoogroups.com )

I'm trying to do frequency detection on a class of devices whose PCM encoder
applies a high pass filter. My frequency detection algorithm works well unless I
try and do frequency detection for the lower notes that are dampened by the HPF.

>
>Determining the frequency with the highest peak should work fine for the
types of sounds I use (the fundamental frequency is always the loudest in my
environment). However, in the case where the fundamental frequency lies within
the dampened region of the HPF, its amplitude shows up on the spectrum as being
much lower than its harmonic frequencies. 
>
>So I tried applying harmonic product spectrum analysis. While there was some
improvement over highest-peak detection, the harmonic frequencies are still
being detected for most sounds with a fundamental frequency in the dampened
region. With HPS, it seems to just detect the first major harmonic outside of
the dampened region.
>
>I've eliminated the possibility that the error in detection is due to
aliasing. 
>
>The HPF seems to lower the amplitudes on a sliding scale by up to 50dB (the
lower the note, the greater the reduction). I've attempted to reverse the HPF by
applying a low pass filter appropriately, but it's too difficult to determine
the amount of amplitude reduction occurring. The HPF used on each device is
slightly different, meaning I need to make a unique LPF for each device, which
is just not feasible.
>
>I would have given up trying to detect the frequency of the lower notes on
these devices, however I see other applications are able to detect lower
frequencies flawlessly. My question is, what frequency detection algorithm could
these other applications using? Any advice?
>
>Thank you,
>Scott

Hi Scott, 

Why not using any of the other techniques for pitch detection, such as
autocorrelation or cepstral analysis?

Other idea, you can apply linear prediction analysis frame by frame in order to
get a whitened version of the signal. So, then you can use HPS or a comb
filter.
- Felipe

______________________________
New Code Sharing Section now Live on DSPRelated.com. Learn about the Reward Program for Contributors here.



(You need to be a member of audiodsp -- send a blank email to audiodsp-subscribe@yahoogroups.com )