Forums

How to mix pcm sample?

Started by galip0986 December 1, 2008
Hello,

I am implementing a voice conferencing program. I am using
tms320vc5416 dsp.I have a distortion problem. I have A-law expanded(13
bit) pcm samples. These samples are stored as unsigned short in
buffers. I convert this samples signed int and then I simply add these
samples together. I abstract users own voice from the sum. At the and
I scale the conclusion(shift right 2 bits). Then I convert signed int
samples to unsigned short samples. Voice quality isn't good. How can I
improve voice quality.

I would appreciate any help in this regard. Thanks a lot, in advance.

Regards,

Galip
Hi,

You probably have saturation problem when you sum the samples.

You could try shifting right the samples before the summation.

Regards,

Einat

From: s... [mailto:s...]
On Behalf Of galip0986
Sent: Monday, December 01, 2008 11:13 AM
To: s...
Subject: [speechcoding] How to mix pcm sample?

Hello,

I am implementing a voice conferencing program. I am using
tms320vc5416 dsp.I have a distortion problem. I have A-law expanded(13
bit) pcm samples. These samples are stored as unsigned short in
buffers. I convert this samples signed int and then I simply add these
samples together. I abstract users own voice from the sum. At the and
I scale the conclusion(shift right 2 bits). Then I convert signed int
samples to unsigned short samples. Voice quality isn't good. How can I
improve voice quality.

I would appreciate any help in this regard. Thanks a lot, in advance.

Regards,

Galip
Hi,

Before converting signed short, you can check saturation. Or earlier you do
it .

Regards,
Pandu.
On Mon, Dec 1, 2008 at 2:43 PM, galip0986 wrote:

> Hello,
>
> I am implementing a voice conferencing program. I am using
> tms320vc5416 dsp.I have a distortion problem. I have A-law expanded(13
> bit) pcm samples. These samples are stored as unsigned short in
> buffers. I convert this samples signed int and then I simply add these
> samples together. I abstract users own voice from the sum. At the and
> I scale the conclusion(shift right 2 bits). Then I convert signed int
> samples to unsigned short samples. Voice quality isn't good. How can I
> improve voice quality.
>
> I would appreciate any help in this regard. Thanks a lot, in advance.
>
> Regards,
>
> Galip
>
>
>

Even more, check which stream is causing saturation , do an gain reduction untill no saturation happens.

On Tue, 02 Dec 2008 Pandu Ranganadh M wrote :
>Hi,
>
>Before converting signed short, you can check saturation. Or earlier you do
>it .
>
>Regards,
>Pandu.
>On Mon, Dec 1, 2008 at 2:43 PM, galip0986 wrote:
>
> > Hello,
> >
> > I am implementing a voice conferencing program. I am using
> > tms320vc5416 dsp.I have a distortion problem. I have A-law expanded(13
> > bit) pcm samples. These samples are stored as unsigned short in
> > buffers. I convert this samples signed int and then I simply add these
> > samples together. I abstract users own voice from the sum. At the and
> > I scale the conclusion(shift right 2 bits). Then I convert signed int
> > samples to unsigned short samples. Voice quality isn't good. How can I
> > improve voice quality.
> >
> > I would appreciate any help in this regard. Thanks a lot, in advance.
> >
> > Regards,
> >
> > Galip
> >
> >
> >
I am not famliar with it. But how can you abstarct signal from the mixed?
5+5,2+8,3+7. Only knowing the sum,how do u know the combined add components?

cheers
hyeewang

Hello,
>
>I am implementing a voice conferencing program. I am using
>tms320vc5416 dsp.I have a distortion problem. I have A-law expanded(13
>bit) pcm samples. These samples are stored as unsigned short in
>buffers. I convert this samples signed int and then I simply add these
>samples together. I abstract users own voice from the sum. At the and
>I scale the conclusion(shift right 2 bits). Then I convert signed int
>samples to unsigned short samples. Voice quality isn't good. How can I
>improve voice quality.
>
>I would appreciate any help in this regard. Thanks a lot, in advance.
>
>Regards,
>
>Galip
>
Hi,
 
Demixing speech out of mixed voices is not easy abstraction.  Use specific pitch filters to seperate out a particular pitch of the signal. Speech is much sensitive model as it all is at a very low frequency. Every human voice has its own pitch and othe parameters.  Go through the encoding process of any speech coder as to understand how can you synthesize and analyze speech.  Here you may not need to do code search or any technique but you need to convert speech into those parameters of the speech (like pitch etc) seperate out the parameters there and then convert back to PCM
 
Prem

;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-
 
Excellence is not an act....but a habit........

Money comes and goes....morality comes and grows......
 
;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;-;--;-;-

--- On Thu, 4/12/08, h...@yahoo.com.cn wrote:

From: h...@yahoo.com.cn
Subject: [speechcoding] Re: How to mix pcm sample?
To: s...
Date: Thursday, 4 December, 2008, 7:47 AM

I am not famliar with it. But how can you abstarct signal from the mixed?
5+5=10,2+8=10, 3+7=10. Only knowing the sum,how do u know the combined add components?

cheers
hyeewang

Hello,
>
>I am implementing a voice conferencing program. I am using
>tms320vc5416 dsp.I have a distortion problem. I have A-law expanded(13
>bit) pcm samples. These samples are stored as unsigned short in
>buffers. I convert this samples signed int and then I simply add these
>samples together. I abstract users own voice from the sum. At the and
>I scale the conclusion(shift right 2 bits). Then I convert signed int
>samples to unsigned short samples. Voice quality isn't good. How can I
>improve voice quality.
>
>I would appreciate any help in this regard. Thanks a lot, in advance.
>
>Regards,
>
>Galip
>----------- --------- --------- -------