Sign in
Search speechcoding
Subscribe to speechcoding
speechcoding by Keywords
ACELP | ADPCM | AMBE | AMR | AMR-NB | CELP | Codebook | DTMF | G.723 | G.726 | G.729 | GSM | Interpolation | LPC | LSF | LSP | MELP | PCM | Perceptual | Pitch | PSOL | QCELP | Quantization | SMV | VAD | Vocoder
Sponsor
Discussion Groups
Discussion Groups | Speech Coding | Re: No Subject
Technical discussions related to Speech Coding (all itu and other vocoders, ACELP, CELP, AMR, etc)
No Subject - Eris Ristemena - Jul 18 5:45:00 1999
|
hello,
is this mail group still work?
thanks.
Eris Ristemena
s...@stttelkom.ac.id
DSP Laboratory
STT Telkom Bandung
62-022-7564184 ext 2326
www.egroups.com - Simplifying group communications |
|
|
(You need to be a member of speechcoding -- send a blank email to speechcoding-subscribe@yahoogroups.com )
Re: No Subject - Stephane Boucher - Jul 18 16:56:00 1999
|
Eris,
The group is still working.
Actually, this is a new group and there is more than 40 (as of 18/07/99)
speech coding "experts" on that list.
Your are all invited to share useful
information or ask speech coding related questions.
Stephane Boucher
Speech Coding Group
University of Sherbrooke
www.egroups.com - Simplifying group communications |
|
|
(You need to be a member of speechcoding -- send a blank email to speechcoding-subscribe@yahoogroups.com )
Re: No Subject - Author Unknown - Jul 18 18:02:00 1999
|
yes work,do you like to join us ! At 12:45 ÚÕÑ 18/07/99 +0700, you wrote: > hello, is this mail group still work? thanks. Eris Ristemena > DSP Laboratory STT Telkom Bandung 62-022-7564184 >ext 2326 > > Click Here! eGroups.com home: >http://www.egroups.com/group/speechcoding > www.egroups.com - Simplifying group communications |
(You need to be a member of speechcoding -- send a blank email to speechcoding-subscribe@yahoogroups.com )
No Subject - Eris Ristemena - Jul 25 22:50:00 1999
|
hi, I want to ask, do i have to get permission from the moderator first before my email can be distributed to all member of this mailing list? thanks Eris |
(You need to be a member of speechcoding -- send a blank email to speechcoding-subscribe@yahoogroups.com )
No Subject - Eris Ristemena - Jul 27 3:18:00 1999
|
----- FLOATING POINT PROBLEM ------ Hi, i'm currently developing an algorithm about speech coding using DSP processor 16 bit fixed point TMS320C25. Because my program need to be operated in floating point number , so i must manipulate the ability of my processor which is build as fixed point processor to be working in floating point format number. The simple format of floating point which i consider to use is Q15 format. The format is like the following (just for clear) : - any decimal number can be representated in Q15 floating point format as one most significant bit (MSB) as sign bit, and another 15 bits as mantissa (number on the left of point). for example : 0,25 can be representated in Q15 format as 0010 0000 0000 0000 or 2000h -0,25 can be representated in Q15 format as 1010 0000 0000 0000 or 2000h the problem is when we operate the Q15 format in arithmetic (sum, substraction, multiply and division) the error (wrong answer) will always occure if one of the operand is negatif, for example : 0,25 + (-0,25) in Q15 format : 2000h + A000h = C000 h (it should be zerro but its not, error occured) the same case will also happen in substraction, multiplication and division. the alternative way that i could think of is to detect both operand (for e.g A and B) and decide wheather both positif, both negative, only A positif, or only B positif. And then each case will be treated in different way, the aim is to avoid any non positif operand. This algorithm, i think, will make my speech coding scheme more complex and slower by the delay process it makes. So, is there any other algorithm used for arithmetic operation of Q15 floating point format which more simple and short, or maybe is there any other trick and treat for this operation??? i will appreciate any suggestion. Thanks Eris ------------------------------------------------------------ Eris Ristemena () Digital Signal Processing Laboratory STT Telkom Bandung Indonesia 62-22-7564108 ext. 2326 ------------------------------------------------------------ |
|
|
(You need to be a member of speechcoding -- send a blank email to speechcoding-subscribe@yahoogroups.com )
Re: No Subject - Author Unknown - Jul 27 16:02:00 1999
|
> i'm currently developing an algorithm about speech coding using DSP > processor 16 bit fixed point TMS320C25. Because my program need to be > operated in floating point number , so i must manipulate the ability of my > processor which is build as fixed point processor to be working in floating > point format number. Eric, why don't you try to implement the LPC10 algorithm in 16-bit fixed point, instead of simulating the floating point arithmetic? Of course it takes longer, as you have to find the appropiate scaling and normalizations to be applied in every point of the algorithm, and also some variables do not hold in 16 bit registers, so you have to find some work-arounds (register concatenation, localized floating-point), but you get a much efficient (faster) implementation at the end. Or even it could made the difference between your algorithm fitting into one TMS320C25 or not. I did some of this work for the implementation of some blocks of the CELPFS1016 (the LPC calculation and quantization)on a DSP56001 . If you like, give a look to chapter 7 of my thesis in: ftp://fox.unine.ch/pub/GRASSI/mathese.pdf or http://www-imt.unine.ch/www/grp_pe/www/people/grassi/personal/welcome.h tml also in chapter 3 and 4 is explained how to simulate quantization effects. BTW, the ETSI standards (GSM06.10, GSM06.20, GSM06.60 and the new AMR) and some of the ITU-T (G.729 and G.723.1) standards are implemented in C code, simulating 16-bit fixed point arithmetic. So if you give a look to any of these programs, you will get ideas on how to do your implementation. Good luck, Sara > The simple format of floating point which i consider to use is Q15 format. > The format is like the following (just for clear) : > - any decimal number can be representated in Q15 floating point format as > one most significant bit (MSB) as sign bit, and another 15 bits as mantissa > (number on the left of point). for example : > 0,25 can be representated in Q15 format as 0010 0000 0000 0000 or 2000h > -0,25 can be representated in Q15 format as 1010 0000 0000 0000 or 2000h > > the problem is when we operate the Q15 format in arithmetic (sum, > substraction, multiply and division) the error (wrong answer) will always > occure if one of the operand is negatif, for example : > 0,25 + (-0,25) in Q15 format : 2000h + A000h = C000 h (it should be zerro > but its not, error occured) > the same case will also happen in substraction, multiplication and division. > > the alternative way that i could think of is to detect both operand (for e.g > A and B) and decide wheather both positif, both negative, only A positif, or > only B positif. And then each case will be treated in different way, the aim > is to avoid any non positif operand. This algorithm, i think, will make my > speech coding scheme more complex and slower by the delay process it makes. > > So, is there any other algorithm used for arithmetic operation of Q15 > floating point format which more simple and short, or maybe is there any > other trick and treat for this operation??? > > i will appreciate any suggestion. > Thanks > Eris > > ------------------------------------------------------------ > Eris Ristemena () > Digital Signal Processing Laboratory > STT Telkom Bandung > Indonesia > 62-22-7564108 ext. 2326 |
(You need to be a member of speechcoding -- send a blank email to speechcoding-subscribe@yahoogroups.com )
No Subject - Author Unknown - Aug 3 1:42:00 1999
|
Is there any one who can tell me something about RCELP- Relaxation CELP. what's the difference between RCELP and the other CELP algorithm? Thanks. ________________________________________________ »¶ÓÄúʹÓÃ163µç×ÓÓʾÖÃâ·Ñ·þÎñ http://www.163.net 163³¬¼¶¿áÈ«¹ú×î´óµÄÃâ·Ñ¸öÈËÖ÷Ò³»ùµØhttp://cool.163.net |
(You need to be a member of speechcoding -- send a blank email to speechcoding-subscribe@yahoogroups.com )