[George[_4_]]

I'm looking for help finding info on very small audio segments in ordinary
speech and music.

Specifically, what are the characteristics of audio segments that tend to
occur most frequently? Segments here refers to durations of under a
millisecond or so. Obviously the most common occurrence would be segments
of silence. But during active audio, what patterns tend to occur more than
others? (For example, a pattern could an increase in composite audio level
of X% over Y time period.)

Reason for this question is that my application will search for these
relatively frequent occurrences in streaming audio and mark them as time
references. I just need to define them.

Thanks for any help.

[Peter Larsen[_2_]]

George wrote:

I'm looking for help finding info on very small audio segments in
ordinary speech and music.

Specifically, what are the characteristics of audio segments that
tend to occur most frequently? Segments here refers to durations of
under a millisecond or so. Obviously the most common occurrence
would be segments of silence. But during active audio, what patterns
tend to occur more than others? (For example, a pattern could an
increase in composite audio level of X% over Y time period.)


Why would silence be the most common short audio item?

Reason for this question is that my application will search for these
relatively frequent occurrences in streaming audio and mark them as
time references. I just need to define them.

It is like looking at the North Sea and wait for it to repeat itself.

Thanks for any help.

You could have a sporting chance with natural audio, but with processed
audio ... nah, I'd keep my day job if I were you.


Kind regards

Peter Larsen

Kind regards

Peter Larsen

[[email protected]]

On Jan 1, 11:12 pm, "George" wrote:
I'm looking for help finding info on very small audio segments in ordinary
speech and music.

Specifically, what are the characteristics of audio segments that tend to
occur most frequently? Segments here refers to durations of under a
millisecond or so. Obviously the most common occurrence would be segments
of silence.

Why do you think this is so? On what kind of material?

But during active audio, what patterns tend to occur more than
others? (For example, a pattern could an increase in composite audio level
of X% over Y time period.)
Reason for this question is that my application will search for these
relatively frequent occurrences in streaming audio and mark them as time
references. I just need to define them.

Unfortunately, you've constrained the problem in such a way
as to make it very difficult to get what you think you want.
If you're limiting the window size to a millisecond, that, all
by itself, will limit you to looking for information over 1 kHz:
nothing below will get through. From a practical implementation
standpoint, your windowing, unless you are very careful,
itself will gen4erate artifacts, and what you may find is the
most common pattern is, in fact, windowing artifacts themselves.

el

[Richard Crowley]

"George" wrote ...
Reason for this question is that my application will search for these
relatively frequent occurrences in streaming audio and mark them as
time references.

Why?

[George[_4_]]

Hope this clarification will help.

Why would silence will be more common than any other type of event? Purely
an assumption. Intersyllabic silence during speech, pauses between music
and such are the events noticeable to the human ear so they seem to occur
often, but I shouldn't generalize that silence is statistically the most
common type of event.

Why limit the segment length? In the interest of minimizing computation
time and complexity and because we want to test for an "easy" match that
occurs frequently.

To explain:

The audio material will be digitized in an ADC then tested for a match with
a pre-defined condition. The audio material from speech or random music
sources is treated as simply a signal source for pattern testing that we
throw away after testing. We want to define patterns that are statistically
likely to occur at a fairly frequent rate of say a few times per second in
streaming audio. Long segments implies more rigorous match criteria and
lower match frequency than we need.

So for example say we test pairs of sample outputs from the ADC and our
match criterion is the second sample must represent higher audio amplitude
than the first. That's way too common. But now say we test segments of
five samples each and require that all five have increasing amplitude.
Statistically how often is that likely to occur?

My question comes down to: where is information like this available?

Thanks for any suggestions.



"George" wrote in message
. net...
I'm looking for help finding info on very small audio segments in ordinary
speech and music.

Specifically, what are the characteristics of audio segments that tend to
occur most frequently? Segments here refers to durations of under a
millisecond or so. Obviously the most common occurrence would be segments
of silence. But during active audio, what patterns tend to occur more
than others? (For example, a pattern could an increase in composite audio
level of X% over Y time period.)

Reason for this question is that my application will search for these
relatively frequent occurrences in streaming audio and mark them as time
references. I just need to define them.

Thanks for any help.

[Peter Larsen[_2_]]

George wrote:

Why would silence will be more common than any other type of event? Purely
an assumption. Intersyllabic silence during speech, pauses
between music and such are the events noticeable to the human ear so
they seem to occur often, but I shouldn't generalize that silence is
statistically the most common type of event.

Intersyllabic silence is less silent than you assume.

My question comes down to: where is information like this available?

Record some audio of the type you want to [whatever it is you want].


Kind regards

Peter Larsen

[Richard Crowley]

"George" wrote ...
Hope this clarification will help.

Why would silence will be more common than any other type of event?

Maybe it would, and maybe it woudn't.
Depends on the source material.

Purely an assumption. Intersyllabic silence during speech, pauses between
music and such are the events noticeable to the human ear

And the sounds inbetween are NOT "noticibale to
the human ear"? Something seems fundamentally
flawed in your presumption.

so they seem to occur often,

Based on what, exactly?

but I shouldn't generalize that silence is statistically the most common
type of event.

I don't think you can even generalize that "the most
common type of event" even exists.

Why limit the segment length? In the interest of minimizing computation
time and complexity and because we want to test for an "easy" match that
occurs frequently.

To explain:

The audio material will be digitized in an ADC then tested for a match
with a pre-defined condition. The audio material from speech or random
music sources is treated as simply a signal source for pattern testing
that we throw away after testing. We want to define patterns that are
statistically likely to occur at a fairly frequent rate of say a few times
per second in streaming audio. Long segments implies more rigorous match
criteria and lower match frequency than we need.

So for example say we test pairs of sample outputs from the ADC and our
match criterion is the second sample must represent higher audio amplitude
than the first. That's way too common. But now say we test segments of
five samples each and require that all five have increasing amplitude.
Statistically how often is that likely to occur?

Depends VERY MUCH on what you are listening to.
For random speech, I'd bet that recurring short sample
duplication is much more rare than you think.

My question comes down to: where is information like this available?

Why are you doing this? Trace back to the prior art or other
published similar experiments. Your questions don't travel
very well as generic experiments.