Re: Downshifted pitches and sample-rates to decrease bandwidth-usage of audio files.

On Sep 11, 8:09 pm, cincy...@gmail.com wrote in 
http://groups.google.com/group/comp.dsp/msg/c947ae2e85185862?hl=en& :

 > As always, what's important is the total bandwidth of the signal.
 > Shifting things around in frequency doesn't buy you anything; your
 > minimum sampling frequency is still determined by the total bandwidth
 > of the signal, which you can't change without distorting it in some
 > way. Basically, all you're describing is shifting some bandpass signal
 > down to (almost) baseband, which doesn't do anything for you with
 > respect to compressibility.

Isn�t the minimum frequency linear-PCM audio can encode 0 Hz, regardless 
of the sample-rate?

If there is no minimum frequency, they why wouldn�t my type of 
downshifting work for digital devices with low-bandwidth/storage-space?

I can understand that it wouldn�t work for a telephone system because 
the telephone lines have not only a maximum-frequency limit but also a 
minimum. Sounds below 300 Hz or above 3,000 Hz will be cut-off by 
devices that process signals on the telephone lines. However, in 
linear-PCM there is not cut-off for the lower-end. You can get as low 
and you want and still get your signal coherently encoded [provided the 
bit-resolution is sufficient.]. In linear-PCM, the cut-off affects 
high-frequencies only -- frequencies that go above what the sample-rate 
allows. This causes aliasing.

On Sep 11, 11:42 pm, "Green Xenon [Radium]" <gluceg...@excite.com>
wrote:
> Isn't the minimum frequency linear-PCM audio can encode 0 Hz, regardless
> of the sample-rate?
>
> If there is no minimum frequency, they why wouldn't my type of
> downshifting work for digital devices with low-bandwidth/storage-space?

If you downshift, the frequency spacing will become smaller.
You may have to increase the duration to resolve the separate
frequencies, which will require more storage proportional to the
downshift.  No free savings.

Ron N. wrote:

> If you downshift, the frequency spacing will become smaller.
> You may have to increase the duration to resolve the separate
> frequencies, which will require more storage proportional to the
> downshift.  

If that were the case, then how can Adobe Audition decrease the 
frequencies of the audio in a file without increasing the length of the 
file?

On Sep 12, 3:33 pm, "Green Xenon [Radium]" <gluceg...@excite.com>
wrote:
> Ron N. wrote:
> > If you downshift, the frequency spacing will become smaller.
> > You may have to increase the duration to resolve the separate
> > frequencies, which will require more storage proportional to the
> > downshift.
>
> If that were the case, then how can Adobe Audition decrease the
> frequencies of the audio in a file without increasing the length of the
> file?

Have you checked to see if you can still resolve
several closely spaced frequencies?  Or does it
throw away or mix together some spectral bins?

Ron N. wrote:

> Have you checked to see if you can still resolve
> several closely spaced frequencies?  Or does it
> throw away or mix together some spectral bins?

Apparently -- from what I see on the graph -- it decreases the amount of 
cycles per distance and then increases the lengths of the waves. So 
basically the signal does not take up more space than it did prior to 
the pitch-shifting. The amount of waves per area decreases but the 
remaining waves are increased in length, so the distance from where the 
waves started to where they finish do not decrease -- their lengths 
increase but their amounts decrease. So overall, the distance from start 
to finish remains the same. Hence the file�s length does not change.

Green Xenon [Radium] wrote:
> Ron N. wrote:
> 
>> If you downshift, the frequency spacing will become smaller.
>> You may have to increase the duration to resolve the separate
>> frequencies, which will require more storage proportional to the
>> downshift.  
> 
> If that were the case, then how can Adobe Audition decrease the 
> frequencies of the audio in a file without increasing the length of the 
> file?

The normal shift is slight. There has to be enough excess sample rate to 
accommodate it. Since the original was sampled at an adequate rate for 
audio, upward shifts work. With a sample rate of 8 KHz, you lose 
everything above 2 KHz if you shift up by an octave. Expecting otherwise 
shows that you weren't thinking clearly.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

On Sep 12, 4:46 pm, "Green Xenon [Radium]" <gluceg...@excite.com>
wrote:
> Ron N. wrote:
> > Have you checked to see if you can still resolve
> > several closely spaced frequencies?  Or does it
> > throw away or mix together some spectral bins?
>
> Apparently -- from what I see on the graph -- it decreases the amount of
> cycles per distance and then increases the lengths of the waves. So
> basically the signal does not take up more space than it did prior to
> the pitch-shifting. The amount of waves per area decreases but the
> remaining waves are increased in length, so the distance from where the
> waves started to where they finish do not decrease -- their lengths
> increase but their amounts decrease. So overall, the distance from start
> to finish remains the same. Hence the file's length does not change.

Yes.  If the particular sample you were looking at
had a sufficient amount of repetition, then its
information content was "low", and the data can thus
be compressed by a variety of means.

What do you do if the information content is higher,
for example there is no exact repetition and every
wave has a slightly different length?  If you decrease
the number of waves, then some of the wave lengths
will end up completely missing.  Does that meet your
lossy quality metric?

Jerry Avins wrote:

> The normal shift is slight. There has to be enough excess sample rate to 
> accommodate it. Since the original was sampled at an adequate rate for 
> audio, upward shifts work. 

AFAIK, upshifting is even more restricted than downshifted. This is 
obvious from what you say below. A 44.1 KHz sample rate can 
mathematically handle a max frequency of 22.05 KHz. Physically, it can 
handle even less. If the frequencies are upshifted beyond what a 44.1 
KHz sample rate allows, then aliasing will occur.

> With a sample rate of 8 KHz, you lose 
> everything above 2 KHz if you shift up by an octave. 

Yes.

Green Xenon [Radium] wrote:
> Jerry Avins wrote:
> 
>> The normal shift is slight. There has to be enough excess sample rate 
>> to accommodate it. Since the original was sampled at an adequate rate 
>> for audio, upward shifts work. 
> 
> AFAIK, upshifting is even more restricted than downshifted. This is 
> obvious from what you say below. A 44.1 KHz sample rate can 
> mathematically handle a max frequency of 22.05 KHz. Physically, it can 
> handle even less. If the frequencies are upshifted beyond what a 44.1 
> KHz sample rate allows, then aliasing will occur.
> 
>> With a sample rate of 8 KHz, you lose everything above 2 KHz if you 
>> shift up by an octave. 

Maybe you get it after all. Now apply that to your unworkable 
compression scheme with numbers and see where you get.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Ron N. wrote:

> On Sep 12, 4:46 pm, "Green Xenon [Radium]" <gluceg...@excite.com>
> wrote:


>>Apparently -- from what I see on the graph -- it decreases the amount of
>>cycles per distance and then increases the lengths of the waves. So
>>basically the signal does not take up more space than it did prior to
>>the pitch-shifting. The amount of waves per area decreases but the
>>remaining waves are increased in length, so the distance from where the
>>waves started to where they finish do not decrease -- their lengths
>>increase but their amounts decrease. So overall, the distance from start
>>to finish remains the same. Hence the file's length does not change.

> Yes.  If the particular sample you were looking at
> had a sufficient amount of repetition, then its
> information content was "low", and the data can thus
> be compressed by a variety of means.

Okay.

> What do you do if the information content is higher,
> for example there is no exact repetition and every
> wave has a slightly different length?  
> If you decrease
> the number of waves, then some of the wave lengths
> will end up completely missing.  Does that meet your
> lossy quality metric?

Those are good questions I wish I could answer. However, for most of my 
songs, pitch-shifting works fine. I can make Backstreet Boys [a band of 
high-pitched singers] sound like Creed or Papa Roach [bands of 
low-pitched singers] without noticeable distortion. I can also make 
Bosson sound like Celine Dion without perceptible artifacts.