Hi, I am doing project named Implementation of Mu-law Speech companding on TMS320vc5510 using DSK5510 board. I need help regarding algorithm for mu-law My data is 16-bit samples that i am getting from reading .WAVE file. I want to write in C language.I am implementing it offline and real time both ways. I neeed help on look-up table, What look-up table i will be required and How can i map 16 to 8 and 8 to 16 bit ? pls send me required material on that or source code of mu-law compression and expansion. or pls give me help on real time implementation of it Vikas Patil IIIT, Pune vikasp_june05@embed.isquareit.ac.in
Mu-Law Algorithm
Started by ●February 5, 2006
Reply by ●February 5, 20062006-02-05
On Sun, 05 Feb 2006 09:00:37 -0600, "vikaspatil123" <vikaspatil19@yahoo.co.in> wrote:>Hi, > > I am doing project named Implementation of Mu-law Speech companding on >TMS320vc5510 using DSK5510 board. > >I need help regarding algorithm for mu-law My data is 16-bit samples that >i am getting from reading .WAVE file. >I want to write in C language.I am implementing it offline and real time >both ways. > >I neeed help on look-up table, What look-up table i will be required and >How can i map 16 to 8 and 8 to 16 bit ? > >pls send me required material on that or source code of mu-law compression >and expansion. or pls give me help on real time implementation of itMu Law is specified in the ITU-T Recommendation G.711. You could try googling for G.711. You will find several implementions that might be useful to you. Some will be free, some will be commercial. Some may even work well. G.711 also specifies A-law, so make sure you don't mix them up. Here is the wikipedia entry: http://en.wikipedia.org/wiki/G.711 Here is the ITU-T Recommendation, which should be regarded as the canonical source of information about mu-law: http://www.itu.int/rec/T-REC-G.711/en Note that the ITU-T have a reference implemention of G.711 written in ANSI C. This page: http://www.itu.int/rec/recommendation.asp?type=products&lang=e&parent=T-REC-G contains this note: "Corresponding ANSI-C code is available in the G.711 module of the ITU-T G.191 Software Tools Library." Regards, Allan
Reply by ●February 5, 20062006-02-05
vikaspatil123 wrote:> Hi, > > I am doing project named Implementation of Mu-law Speech companding on > TMS320vc5510 using DSK5510 board. > > I need help regarding algorithm for mu-law My data is 16-bit samples that > i am getting from reading .WAVE file. > I want to write in C language.I am implementing it offline and real time > both ways. > > I neeed help on look-up table, What look-up table i will be required and > How can i map 16 to 8 and 8 to 16 bit ? > > pls send me required material on that or source code of mu-law compression > and expansion. or pls give me help on real time implementation of it > > > Vikas Patil > IIIT, Pune > vikasp_june05@embed.isquareit.ac.in > > > >Mu law maps a 12 (or 13) bit number to 8 bits, so it is, perforce, a 4096 to 256 mapping (or 4096 to 255, or 8192 to 256 -- look at the standard). As such it is lossy, and when you go back to 12 (or 13) bits there will be gaps -- they'll just be gaps that don't affect speech quality too much. In addition to being lossy, mu law compandors don't have much dynamic range -- the 12 bits that you start from should be pretty well filled with data. That means that you should apply some AVC (automatic volume control) to the data before it hits your compandor. If you don't you'll either be clipping or sending data that's too quiet. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com
Reply by ●February 5, 20062006-02-05
From my notes: -Clark
-----------------------------------------------------------------------
TWIMC,
Below are two routines I wrote for converting between ulaw and linear.
I use them with the SparcStation and they seem to work fine.
I am pretty sure (99.9%) that they implement the standard as specified
in the references.
Note that the standard deals with converting between 12 bit linear
and 8 bit ulaw. These routines assume 16 bit linear. Thus, some
bit shifting may be necessary.
craig
------------------------------------------------------------------
/**
** Signal conversion routines for use with the Sun4/60 audio chip
**/
/*
* This routine converts from linear to ulaw
* 29 September 1989
*
* Craig Reese: IDA/Supercomputing Research Center
* Joe Campbell: Department of Defense
*
* References:
* 1) CCITT Recommendation G.711 (very difficult to follow)
* 2) "A New Digital Technique for Implementation of Any
* Continuous PCM Companding Law," Villeret, Michel,
* et al. 1973 IEEE Int. Conf. on Communications, Vol 1,
* 1973, pg. 11.12-11.17
* 3) MIL-STD-188-113,"Interoperability and Performance Standards
* for Analog-to_Digital Conversion Techniques,"
* 17 February 1987
*
* Input: Signed 16 bit linear sample
* Output: 8 bit ulaw sample
*/
#define ZEROTRAP /* turn on the trap as per the MIL-STD */
#define BIAS 0x84 /* define the add-in bias for 16 bit samples */
#define CLIP 32635
unsigned char
linear2ulaw(sample)
int sample;
{
static int exp_lut[256] = {0,0,1,1,2,2,2,2,3,3,3,3,3,3,3,3,
4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,
5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7};
int sign, exponent, mantissa;
unsigned char ulawbyte;
/** get the sample into sign-magnitude **/
sign = (sample >> 8) & 0x80; /* set aside the sign */
if (sign != 0) sample = -sample; /* get magnitude */
if (sample > CLIP) sample = CLIP; /* clip the magnitude */
/** convert from 16 bit linear to ulaw **/
sample = sample + BIAS;
exponent = exp_lut[(sample>>7) & 0xFF];
mantissa = (sample >> (exponent+3)) & 0x0F;
ulawbyte = ~(sign | (exponent << 4) | mantissa);
#ifdef ZEROTRAP
if (ulawbyte == 0 ) ulawbyte = 0x02; /* optional CCITT trap */
#endif
/** return the result **/
return(ulawbyte);
}
/*
* This routine converts from ulaw to 16 bit linear
* 29 September 1989
*
* Craig Reese: IDA/Supercomputing Research Center
*
* References:
* 1) CCITT Recommendation G.711 (very difficult to follow)
* 2) MIL-STD-188-113,"Interoperability and Performance Standards
* for Analog-to_Digital Conversion Techniques,"
* 17 February 1987
*
* Input: 8 bit ulaw sample
* Output: signed 16 bit linear sample
*/
int
ulaw2linear(ulawbyte)
unsigned char ulawbyte;
{
static int exp_lut[8]={0,132,396,924,1980,4092,8316,16764};
int sign, exponent, mantissa, sample;
ulawbyte = ~ulawbyte;
sign = (ulawbyte & 0x80);
exponent = (ulawbyte >> 4) & 0x07;
mantissa = ulawbyte & 0x0F;
sample = exp_lut[exponent] + (mantissa << (exponent+3));
if (sign != 0) sample = -sample;
return(sample);
}
-----------------
Craig F. Reese Email: cfreese@super.org
Institute for Defense Analyses/
Supercomputing Research Center
17100 Science Dr.
Bowie, MD 20715-4300
"vikaspatil123" <vikaspatil19@yahoo.co.in> wrote in message
news:7NCdnetS95OIjnvenZ2dnUVZ_sednZ2d@giganews.com...
> Hi,
>
> I am doing project named Implementation of Mu-law Speech companding on
> TMS320vc5510 using DSK5510 board.
>
> I need help regarding algorithm for mu-law My data is 16-bit samples that
> i am getting from reading .WAVE file.
> I want to write in C language.I am implementing it offline and real time
> both ways.
>
> I neeed help on look-up table, What look-up table i will be required and
> How can i map 16 to 8 and 8 to 16 bit ?
>
> pls send me required material on that or source code of mu-law compression
> and expansion. or pls give me help on real time implementation of it
>
>
> Vikas Patil
> IIIT, Pune
> vikasp_june05@embed.isquareit.ac.in
>
>
>
>
Reply by ●February 5, 20062006-02-05
Tim Wescott wrote:> Mu law maps a 12 (or 13) bit number to 8 bits, so it is, perforce, a > 4096 to 256 mapping (or 4096 to 255, or 8192 to 256 -- look at the > standard). As such it is lossy, and when you go back to 12 (or 13) bits > there will be gaps -- they'll just be gaps that don't affect speech > quality too much. > > In addition to being lossy, mu law compandors don't have much dynamic > range -- the 12 bits that you start from should be pretty well filled > with data. That means that you should apply some AVC (automatic volume > control) to the data before it hits your compandor. If you don't you'll > either be clipping or sending data that's too quiet. >The A-law and u-Law are *not* the optimal quantizers for the speech signal statistics. The design point of the u-law and A-law was to have the signal/quantization noise ratio relatively independent of the rms value of the signal. With the addition of the AVC, the simple linear 8 bit quantizer will perform considerably better. What I don't understand is why didn't they apply the fixed preemphasis before quantization. For the voice band signals, they could probably gain another 10dB in SNR. Vladimir Vassilevsky DSP and Mixed-Up Signal Design Consultant http://www.abvolt.com
Reply by ●February 5, 20062006-02-05
Vladimir Vassilevsky wrote:> > > Tim Wescott wrote: > > >> Mu law maps a 12 (or 13) bit number to 8 bits, so it is, perforce, a >> 4096 to 256 mapping (or 4096 to 255, or 8192 to 256 -- look at the >> standard). As such it is lossy, and when you go back to 12 (or 13) >> bits there will be gaps -- they'll just be gaps that don't affect >> speech quality too much. >> >> In addition to being lossy, mu law compandors don't have much dynamic >> range -- the 12 bits that you start from should be pretty well filled >> with data. That means that you should apply some AVC (automatic volume >> control) to the data before it hits your compandor. If you don't >> you'll either be clipping or sending data that's too quiet. >> > > The A-law and u-Law are *not* the optimal quantizers for the speech > signal statistics. The design point of the u-law and A-law was to have > the signal/quantization noise ratio relatively independent of the rms > value of the signal. With the addition of the AVC, the simple linear 8 > bit quantizer will perform considerably better. > What I don't understand is why didn't they apply the fixed preemphasis > before quantization. For the voice band signals, they could probably > gain another 10dB in SNR. > > Vladimir Vassilevsky > > DSP and Mixed-Up Signal Design Consultant > > http://www.abvolt.comGiven that one runs smack into a host of human factors issues when you try to define a cost function I doubt that an 'optimal' quantizer for speech signal statistics will ever be known (or at least never agreed upon). For the state of the art at the time the mu-law and a-law compandors are pretty good. I also couldn't say why they have no preemphasis -- the phone companies were certainly technically astute enough at the time, and willing to invest it some pretty heavy duty hardware. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com
Reply by ●February 5, 20062006-02-05
Vladimir Vassilevsky wrote:> > > Tim Wescott wrote: > > >> Mu law maps a 12 (or 13) bit number to 8 bits, so it is, perforce, a >> 4096 to 256 mapping (or 4096 to 255, or 8192 to 256 -- look at the >> standard). As such it is lossy, and when you go back to 12 (or 13) >> bits there will be gaps -- they'll just be gaps that don't affect >> speech quality too much. >> >> In addition to being lossy, mu law compandors don't have much dynamic >> range -- the 12 bits that you start from should be pretty well filled >> with data. That means that you should apply some AVC (automatic volume >> control) to the data before it hits your compandor. If you don't >> you'll either be clipping or sending data that's too quiet. >> > > The A-law and u-Law are *not* the optimal quantizers for the speech > signal statistics. The design point of the u-law and A-law was to have > the signal/quantization noise ratio relatively independent of the rms > value of the signal. With the addition of the AVC, the simple linear 8 > bit quantizer will perform considerably better. > What I don't understand is why didn't they apply the fixed preemphasis > before quantization. For the voice band signals, they could probably > gain another 10dB in SNR.You get a fairly constant 30dB or so of SNR, which is good enough for most telephony purposes. There isn't a strong advantage is getting that extra 10dB, and the 72dB of overall dynamic range uLaw offers is also perfectly adequate for most telephony purposes. However, considering the original T1 systems were implemented in the very early days of discrete transistors, it is a little surprising they did nothing like you suggest. A small filtering circuit might have eased the complexity of the digital stuff a bit, and every little bit cost serious money back then. Maybe it was foresight - emphasis and deemphasis would have made voice band modems much tougher :-) Regards, Steve
Reply by ●February 5, 20062006-02-05
Vladimir Vassilevsky wrote:> What I don't understand is why didn't they apply the fixed preemphasis > before quantization. For the voice band signals, they could probably > gain another 10dB in SNR.Vladimir, that what happens when a hardware engineer like you is trying to pretend he's a DSP or speech processing expert... The quantization is indeed not optimal to speech since it was not aimed to be so.... ITU-T designed it to encode various signals such as DTMF, fax, music, speech and perhaps other audio signals that may be transmitted on telephony network... Logarithmic quantization does all that pretty well, and hence mu-law, A-law are used in telephony!
Reply by ●February 6, 20062006-02-06
in article d7zFf.12452$H43.6390@trnddc08, DFG at DFG@JustNoSpam.com wrote on 02/05/2006 22:11:> Vladimir Vassilevsky wrote: > >> What I don't understand is why didn't they apply the fixed preemphasis >> before quantization. For the voice band signals, they could probably >> gain another 10dB in SNR. > > Vladimir, that what happens when a hardware engineer like you is trying > to pretend he's a DSP or speech processing expert...dunno what you mean. i thought he hit it correctly with:>> The design point of the u-law and A-law was to have the signal/quantization >> noise ratio relatively independent of the rms value of the signal.but i don't understand what "fixed pre-emphasis" he means. like a fixed boost of the high frequencies similar to the RIAA for vinyl records? -- r b-j rbj@audioimagination.com "Imagination is more important than knowledge."
Reply by ●February 6, 20062006-02-06
robert bristow-johnson wrote:>>>What I don't understand is why didn't they apply the fixed preemphasis >>>before quantization. For the voice band signals, they could probably >>>gain another 10dB in SNR. >>Robert, there is no point to argue with the offended individual. It only makes difficult to the other folks to distinguish you and him.> > but i don't understand what "fixed pre-emphasis" he means. like a fixed > boost of the high frequencies similar to the RIAA for vinyl records?Precisely. The differentiation of the signal to compensate for the spectral tilt before quantization makes noticeable improvement in the quality. This is basically the same idea as the DPCM or RIAA or FM biasing. However AT&T probably decided that 30dB is good enough, especially for that old days. Vladimir Vassilevsky DSP and Mixed-Up Signal Design Consultant http://www.abvolt.com
