FPGA Central

Our goal is to improve the listenability of the speech
signal by decreasing the background noise, without
affecting the intelligibility of the speech.

My question is :
What is the difference between the meaning of "listenability" and
"intelligibility" ?
How to evaluate them?
I am more interested in the concept and method of "listenability" !

Cheers
HyeeWang

On Mar 6, 10:01*pm, HyeeWang <hyeew...@gmail.com> wrote:
> Our goal is to improve the listenability of the speech
> signal by decreasing the background noise, without
> affecting the intelligibility of the speech.
>
> My question is :
> What is the difference between the meaning of "listenability" and
> "intelligibility" ?
> How to evaluate them?
> I am more interested in the concept and method of "listenability" !
>
> Cheers
> HyeeWang

The two are not necessarily closely related. Generally think
"quality"(also pretty vague) VS "understanding", but ...

take a little time and figure out what you mean by "intelligibility".
A word spoken in context is probably alot more "intelligible" than the
same word spoken in isolation. A word spoken from a 10 word
vocabulary is probably a lot more "intelligible" than the same word
spoken as part of a 100 word vocabulary. Combinations of noise,
speech, and listening time in that environment will effect what you
would call intelligibility. Listening time can either increase or
decrease the intelligibility depending on the situation.

Simple general question. Not a simple general answer. You need to do
some research. Figure out what is meaningful to your situation.

On 7 Mar, 04:01, HyeeWang <hyeew...@gmail.com> wrote:
> Our goal is to improve the listenability of the speech
> signal by decreasing the background noise, without
> affecting the intelligibility of the speech.
>
> My question is :
> What is the difference between the meaning of "listenability" and
> "intelligibility" ?

You should ask whoever specified the tasks for your project.

> How to evaluate them?

This is not easy, as this requires you to tailor the
properties of your signal to the inner workings of the
human auditory system. You need to know what properties
of a signal makes it more or less 'intelligable' to the
human mind, quantify these properties, and characterize
the signals with respect to these properties. Not at all
easy.

In fact, the worst-case scenario is that you have to
reproduce most of the human brain in terms of DSP...

Rune

On Mar 7, 5:14*am, Rune Allnor <all...@tele.ntnu.no> wrote:
> On 7 Mar, 04:01, HyeeWang <hyeew...@gmail.com> wrote:
>
> > Our goal is to improve the listenability of the speech
> > signal by decreasing the background noise, without
> > affecting the intelligibility of the speech.
>
> > My question is :
> > What is the difference between the meaning of "listenability" and
> > "intelligibility" ?
>
> You should ask whoever specified the tasks for your project.
>
> > How to evaluate them?
>
> This is not easy, as this requires you to tailor the
> properties of your signal to the inner workings of the
> human auditory system. You need to know what properties
> of a signal makes it more or less 'intelligable' to the
> human mind, quantify these properties, and characterize
> the signals with respect to these properties. Not at all
> easy.
>
> In fact, the worst-case scenario is that you have to
> reproduce most of the human brain in terms of DSP...
>
> Rune

All conventional methods (such as spectral subtraction) have
artifacts, especially in extremely low-snr conditions. So you have to
balance the fact that you have done some magic to eliminate the
traffic noise that surrounds you against the strange "phasey" effects
of the algorithm. It also depends on the end application; if you are a
member of a large conference call, no one wants to hear the summation
of everyones local noise; it can add up pretty quickly, and you would
be willing to trade off "natural" voice quality to get rid of the
noise.

Once you have played with these algorithms for a while, you realize
just how good your ear/brain is at pulling a voice signal out the
noise. Apparently all the cave-men who were poor at this didn't hear
the "look-out! warning and were run over by the Flintstone car, thus
ending their genetic line.

Bob Adams

"HyeeWang" <[email protected]> wrote in message
news:[email protected]...
> Our goal is to improve the listenability of the speech
> signal by decreasing the background noise, without
> affecting the intelligibility of the speech.
>
> My question is :
> What is the difference between the meaning of "listenability" and
> "intelligibility" ?

Listenability = MOS
Intellegibility = DRT

> How to evaluate them?

ITU-T P.8xx recommendations.

Vladimir Vassilevsky
DSP and Mixed Signal Consultant
www.abvolt.com

>On Mar 7, 5:14=A0am, Rune Allnor <all...@tele.ntnu.no> wrote:
>> On 7 Mar, 04:01, HyeeWang <hyeew...@gmail.com> wrote:
>>
>> > Our goal is to improve the listenability of the speech
>> > signal by decreasing the background noise, without
>> > affecting the intelligibility of the speech.
>>
>> > My question is :
>> > What is the difference between the meaning of "listenability" and
>> > "intelligibility" ?
>>
>> You should ask whoever specified the tasks for your project.
>>
>> > How to evaluate them?
>>
>> This is not easy, as this requires you to tailor the
>> properties of your signal to the inner workings of the
>> human auditory system. You need to know what properties
>> of a signal makes it more or less 'intelligable' to the
>> human mind, quantify these properties, and characterize
>> the signals with respect to these properties. Not at all
>> easy.
>>
>> In fact, the worst-case scenario is that you have to
>> reproduce most of the human brain in terms of DSP...
>>
>> Rune
>
>All conventional methods (such as spectral subtraction) have
>artifacts, especially in extremely low-snr conditions. So you have to
>balance the fact that you have done some magic to eliminate the
>traffic noise that surrounds you against the strange "phasey" effects
>of the algorithm. It also depends on the end application; if you are a
>member of a large conference call, no one wants to hear the summation
>of everyones local noise; it can add up pretty quickly, and you would
>be willing to trade off "natural" voice quality to get rid of the
>noise.
>
>Once you have played with these algorithms for a while, you realize
>just how good your ear/brain is at pulling a voice signal out the
>noise. Apparently all the cave-men who were poor at this didn't hear
>the "look-out! warning and were run over by the Flintstone car, thus
>ending their genetic line.

If you know more than one language, you may also find somethin
interesting when you try some serious background noise testing. It is no
surprising that you can pick out your mother tongue accurately under poo
conditions, but need a much clearer signal for languages you know les
well. However, when you try tests you might be surprised how extreme th
difference is. The language (or languages) you learn from a baby seem t
get programmed deep into the way your brain analyses signal, in a wa
languages learned later seldom do.

Regards,
Steve

steveu wrote:


> If you know more than one language, you may also find something
> interesting when you try some serious background noise testing. It is not
> surprising that you can pick out your mother tongue accurately under poor
> conditions, but need a much clearer signal for languages you know less
> well. However, when you try tests you might be surprised how extreme the
> difference is. The language (or languages) you learn from a baby seem to
> get programmed deep into the way your brain analyses signal, in a way
> languages learned later seldom do.

This is true. It is very difficult and fatiguing to communicate in the
second language if the signal is noisy or distorted. I have to focus all
attention. One more detail: for equal conditions, female voice is
recognized much better then the male voice. That's the difference
between the hardware and the software voice decoding.

Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com