similar to: [PATCH] compute_weighted_codebook a little bit faster

Hi, I am bring SPEEX up on Analog Device's Blackfin (using Visual DSP++ tool set). The floating point was too slow on this fixed-point processor so I am going for the integer version. The floating point library build was very clean, but when I build with FIXED_POINT defined, the compiler exits when it find this discrepancy in \libspeex\cb_search.c : ".\libspeex\cb_search.c",

Hi, Jean-Marc Valin wrote: > Well, you could change the order in the encoder as long as you reverse > it in the decoder as well. Ok, I see that in the split_cb_shape_sign_unquant function, that each coefficient is tied to it's position in the nb_subvect exc coefficients. Honestly, I have problems understanding what exactly the codebook search works like. If you have the time to to

I am using speex on a Philips LPC2138 Arm 7 processor and have hit two compiler warnings when using FIXED_POINT and the Arm 4 asm code. In cb_search.c get a warning on the call to compute_weighted_codebook of: "passing arg 2 of 'compute_weighted_codebook' from incompatible pointer type' In cp_search_arm4.h it is defined as: static void compute_weighted_codebook(const signed

Hello, Is it possible to adjust the level of the denoiser ? In an old beta, before you change the aggressiveness of the denoiser, it works very well with the noise of a car, a road etc... but not now. The AEC too works very well in a old beta, but now, I think there is a problem... I have tested it with the same sound card (WB 16000). In any case, your project is the best VOIP library in this

Hi, Jean-Marc Valin wrote: >> 1. What is the point in multiplying a codebook index with some number >> ant adding a loop variable to it as done in the exc unquant function. >> for (j=0;j<subvect_size;j++) >> exc[subvect_size*i+j]+=s*0.03125*shape_cb[ind[i]*subvect_size+j]; > > That's just how you represent a 2D array in C: the codebook has

hi, i am trying to compile a fixed point version of lipspeex on desktop windows environment. a short reason for the complicated story is that we need to keep the 'engine' part of our project same across the various ports of our voip project (blame the pointy haired boss). now, when i compile with FIXED_POINT defined in the Visual C++ 6.0, i get these errors:

Hello, The denoiser was good with a car noise or a raod noise before the r11739. But for some situation, the actual version is perfect. That's why I think that if we can adjust the agressivity of the denoiser, we can respond to all of the situations. Maybe two parameters should be enough : high or low. For the AEC, I have to test with older versions because it's another person that

Hi, Jean-Marc Valin wrote: > Not sure I understand your question. Change the order of what within what? In cb_search, we iterate nb_subvect times over a codebook, finding nb_subvect codebook entries to quantize the excitation signal. After finding these nb_subvect codebook entries, they're written into the stream: /*save indices*/ for (i=0;i<nb_subvect;i++) {

Most asm statements clobber ASTAT bits (shifts, maxes, etc...) but do declare the register as clobbered. Same thing with CC in a few places. Some places make an attempt at clobbering some hardware loop registers, but it's very incomplete compared with how many asm statements actually use hardware loops. Signed-off-by: Mike Frysinger <vapier at gentoo.org> --- libspeex/bfin.h

Hi, I am using speex 1.2beta2 on a 24bit DSP that has a severe program and data space limitations. I am only interested in the speex decoder for 11kbps narrowband implementation. I am using the following parameters and structures 160, /*frameSize*/ 40, /*subframeSize*/ 10, /*lpcSize*/ 17, /*pitchStart*/ 144, /*pitchEnd*/ /* 11 kbps medium bit-rate

Jean-Marc Valin wrote: >Hi Jamey, > >Really cool to see Speex being ported to the C55xx and I'd be glad to >integrate the changes required in Speex (and the style's fine with me). > > I have the encoder and decoder running now and have verified that the encoder is bit-exact wrt to the fixed-point code running on x86 for the same 30-second audio sample. Encode and

Dear Valin, Please, can you explain me the mean of the routines' parameters: 1. void vq_nbest(); 2. static void compute_weighted_codebook(); 3. static inline void target_update(); 4. void split_cb_search_shape_sign(); 5. void split_cb_shape_sign_unquant(); 6. void noise_codebook_quant(); 7. void noise_codebook_unquant(); and what they do briefly??? It's very difficult understand a

Ok, I got a strange case.. I got a bugreport from a user about my application crashing on him. After a bit of trouble, I got him to report back the module and offset it crashed at (win32). I always retain a copy of my symbol files for my release builds, so tracking down the crash address leads to: (gdb) list *0x6d5c2213 0x6d5c2213 is in split_cb_search_shape_sign (cb_search.c:414). 409

> I have the encoder and decoder running now and have verified that the > encoder is bit-exact wrt to the fixed-point code running on x86 for the > same 30-second audio sample. Encode and decode together run in > real-time for 8KHz data, complexity=3, on 120MHz C5509 when code and > data are all in on-chip SRAM. I have not tested the wideband codec yet. Cool! Just curious,

On Mon, 2006-01-23 at 10:15 -0500, tom abcd wrote: > I am using speex on a Philips LPC2138 Arm 7 processor and have hit two > compiler warnings when using FIXED_POINT and the Arm 4 asm code. > > In cb_search.c get a warning on the call to compute_weighted_codebook of: > "passing arg 2 of 'compute_weighted_codebook' from incompatible pointer type' OK, I forgot to

> 1. What is the point in multiplying a codebook index with some number > ant adding a loop variable to it as done in the exc unquant function. > for (j=0;j<subvect_size;j++) > exc[subvect_size*i+j]+=s*0.03125*shape_cb[ind[i]*subvect_size+j]; That's just how you represent a 2D array in C: the codebook has shape_cb_size entries and subvect_size samples per

How about this: the codebook search gives you the N best (in order) combinations of entries for the current sub-frame. You can simply compute a one-bit checksum on all the codebook entries and choose which of the N-best to use based on that and the "message" you have. Of course, there will be cases where all of the N-best match to the same checksum, but that's what error-correcting

Hi, After reading the paper entitled "The Computation of Line Spectral Frequencies Using Chebyshev Polynomials", P. Kabal and R. Ramachandran, IEEE Trans. on ASSP, Vol. 34, No. 6, December 1986, I rewrite the function cheb_poly_eva in lsp.c using the Clenshaw's recurrence formula, as described, for example, in Numerical Recipes in C, Second Edition (5.5 and 5.8) : static float

On Fri, 7 Feb 2014, Timothy B. Terriberry wrote: > Martin Storsjo wrote: >> This is required in order to build using the built-in assembler >> in clang. > > These patches break the gcc build (with "Error: bad instruction"). Ah, right, sorry about that. > Documentation I've seen is contradictory on which order ({cond}{size} or > {size}{cond}) is correct.

I'm trying to make a design decision between a TI 6416 or DM642 (fixed point) and 6713 (floating point) platform. The application is a 32 channel speech encoder. (CBR only, 8khz, 8kbps) To get a feel for the computational load, I am running 1 second (50 frames) of voice through the encoder. My profile of the 6416 indicates I'm at 27.4M cycles/channel. I need to get below 720Mhz/32