July 2009 - FFmpeg-soc - lists.ffmpeg.org

[soc]: r4852 - in amr: amrnbdata.h amrnbdec.c
by cmcq 31 Jul '09

31 Jul '09

Author: cmcq Date: Fri Jul 31 03:47:27 2009 New Revision: 4852 Log: Adjust the quantized LSFs to comply with ref decoder Modified: amr/amrnbdata.h amr/amrnbdec.c Modified: amr/amrnbdata.h ============================================================================== --- amr/amrnbdata.h Fri Jul 31 03:45:05 2009 (r4851) +++ amr/amrnbdata.h Fri Jul 31 03:47:27 2009 (r4852) @@ -2038,6 +2038,7 @@ static const float pred_fac[LP_FILTER_OR #define PRED_FAC_MODE_122 0.65 #define FREQ_LSP_FAC 2.0*M_PI/8000.0 #define FREQ_LSF 1/8000.0 +#define MIN_LSF_SPACING 50.0488 // pitch tables Modified: amr/amrnbdec.c ============================================================================== --- amr/amrnbdec.c Fri Jul 31 03:45:05 2009 (r4851) +++ amr/amrnbdec.c Fri Jul 31 03:47:27 2009 (r4852) @@ -203,6 +203,20 @@ static void interpolate_lsf(float lsf_q[ } /** + * Adjust the quantized LSFs so they are increasing and not too close. + * + * This step isn't mentioned in the spec but is in the reference C decoder. + * Omitting this step creates audible distortion on the sinusoidal sweep + * test vectors in 3GPP TS 26.074. + */ +void adjust_lsfs(float *m) { + int i; + float tmp=0.0; + for (i = 0; i < LP_FILTER_ORDER; i++) + tmp = m[i] = FFMAX(m[i], tmp + MIN_LSF_SPACING); +} + +/** * Decode a set of 5 split-matrix quantized lsf indexes into an lsp vector. * * @param p the context @@ -237,6 +251,8 @@ static void lsf2lsp_for_mode122(AMRConte for (i = 0; i < LP_FILTER_ORDER; i++) lsf[i] += prev_lsf[i]; + adjust_lsfs(lsf); + // store LSF vector for fixed gain smoothing if (update_prev_lsf_r) interpolate_lsf(p->lsf_q, lsf); @@ -299,6 +315,8 @@ static void lsf2lsp_3(AMRContext *p) for (i = 0; i < LP_FILTER_ORDER; i++) lsf_q[i] = lsf_r[i] + p->prev_lsf_r[i] * pred_fac[i] + lsf_3_mean[i]; + adjust_lsfs(lsf_q); + // store LSF vector for fixed gain smoothing interpolate_lsf(p->lsf_q, lsf_q);

3 2

[soc]: r4861 - amr/amrnbdec.c
by cmcq 31 Jul '09

31 Jul '09

Author: cmcq Date: Fri Jul 31 14:50:49 2009 New Revision: 4861 Log: Cosmetics: use consistent naming, use K&R, better spelling, use AMR_BLOCK_SIZE Modified: amr/amrnbdec.c Modified: amr/amrnbdec.c ============================================================================== --- amr/amrnbdec.c Fri Jul 31 14:47:55 2009 (r4860) +++ amr/amrnbdec.c Fri Jul 31 14:50:49 2009 (r4861) @@ -216,7 +216,8 @@ static void interpolate_lsf(float lsf_q[ * Omitting this step creates audible distortion on the sinusoidal sweep * test vectors in 3GPP TS 26.074. */ -static void adjust_lsfs(float *m) { +static void adjust_lsf(float *m) +{ int i; float tmp=0.0; for (i = 0; i < LP_FILTER_ORDER; i++) @@ -258,7 +259,7 @@ static void lsf2lsp_for_mode122(AMRConte for (i = 0; i < LP_FILTER_ORDER; i++) lsf[i] += prev_lsf[i]; - adjust_lsfs(lsf); + adjust_lsf(lsf); // store LSF vector for fixed gain smoothing if (update_prev_lsf_r) @@ -322,7 +323,7 @@ static void lsf2lsp_3(AMRContext *p) for (i = 0; i < LP_FILTER_ORDER; i++) lsf_q[i] = lsf_r[i] + p->prev_lsf_r[i] * pred_fac[i] + lsf_3_mean[i]; - adjust_lsfs(lsf_q); + adjust_lsf(lsf_q); // store LSF vector for fixed gain smoothing interpolate_lsf(p->lsf_q, lsf_q); @@ -1114,7 +1115,7 @@ static int amrnb_decode_frame(AVCodecCon float fixed_gain_factor; float exc_feedback[AMR_SUBFRAME_SIZE]; float fixed_vector[AMR_SUBFRAME_SIZE]; // algebraic code book (fixed) vector - float spare_vector[AMR_SUBFRAME_SIZE]; // extra stack space to hold result from anti-spareness processing + float spare_vector[AMR_SUBFRAME_SIZE]; // extra stack space to hold result from anti-sparseness processing float synth_fixed_gain; // the fixed gain that synthesis should use float *synth_fixed_vector; // pointer to the fixed vector that synthesis should use @@ -1187,7 +1188,7 @@ static int amrnb_decode_frame(AVCodecCon update_state(p); } - ff_acelp_high_pass_filterf(buf_out, p->high_pass_mem, 160); + ff_acelp_high_pass_filterf(buf_out, p->high_pass_mem, AMR_BLOCK_SIZE); for (i = 0; i < AMR_BLOCK_SIZE; i++) // Post-processing up-scales by 2. It's convenient to

1 0

[soc]: r4860 - amr/amrnbdec.c
by cmcq 31 Jul '09

31 Jul '09

Author: cmcq Date: Fri Jul 31 14:47:55 2009 New Revision: 4860 Log: Adjust decoding to correspond to the spec; I can't see why this line was here Modified: amr/amrnbdec.c Modified: amr/amrnbdec.c ============================================================================== --- amr/amrnbdec.c Fri Jul 31 14:43:11 2009 (r4859) +++ amr/amrnbdec.c Fri Jul 31 14:47:55 2009 (r4860) @@ -154,7 +154,6 @@ static enum Mode decode_bitstream(AMRCon if (mode == MODE_DTX) { p->cur_frame_type = RX_SID_FIRST; // get SID type bit - skip_bits(&p->gb, 4); // skip to the next byte if (get_bits1(&p->gb)) // use the update if there is one p->cur_frame_type = RX_SID_UPDATE; /* RFC4867 specifies AMR IF1 from 3GPP TS 26.101 which reverses

1 0

[soc]: r4859 - in amr: amrnbdata.h amrnbdec.c
by cmcq 31 Jul '09

31 Jul '09

Author: cmcq Date: Fri Jul 31 14:43:11 2009 New Revision: 4859 Log: Comments: add references for frame decoding and note a PacketVideo infelicity Modified: amr/amrnbdata.h amr/amrnbdec.c Modified: amr/amrnbdata.h ============================================================================== --- amr/amrnbdata.h Fri Jul 31 06:43:12 2009 (r4858) +++ amr/amrnbdata.h Fri Jul 31 14:43:11 2009 (r4859) @@ -112,7 +112,7 @@ typedef struct { // The following order* tables are used to reorder the bitstream into the bit // allocation format as in the specification. The bits are grouped into three -// classes of importance. See the specification for more information. +// classes of importance. See 3GPP TS 26.101 for more information. #define AMR_LSF(variable, bit) {offsetof(AMRNBFrame, lsf[variable]) >> 1, bit} Modified: amr/amrnbdec.c ============================================================================== --- amr/amrnbdec.c Fri Jul 31 06:43:12 2009 (r4858) +++ amr/amrnbdec.c Fri Jul 31 14:43:11 2009 (r4859) @@ -118,7 +118,9 @@ static av_cold int amrnb_decode_init(AVC /** - * Decode the bitstream into the AMR parameters and discover the frame mode. + * Decode an RFC4867 speech frame into the AMR frame mode and parameters. + * + * The order of speech bits is specified by 3GPP TS 26.101. * * @param p the context * @param buf pointer to the input buffer @@ -155,7 +157,13 @@ static enum Mode decode_bitstream(AMRCon skip_bits(&p->gb, 4); // skip to the next byte if (get_bits1(&p->gb)) // use the update if there is one p->cur_frame_type = RX_SID_UPDATE; - *speech_mode = get_bits(&p->gb, 3); // speech mode indicator + /* RFC4867 specifies AMR IF1 from 3GPP TS 26.101 which reverses + * the Mode Indication bits. This is also used by the reference + * decoder. + * But PacketVideo uses IF2 - it stores these bits the other way + * round. The sample "sound.amr" uses IF2. + */ + *speech_mode = get_bits(&p->gb, 3); // IF2 speech mode indicator } else p->cur_frame_type = RX_SPEECH_GOOD; } else if (mode == NO_DATA) { @@ -1103,7 +1111,7 @@ static int amrnb_decode_frame(AVCodecCon int buf_size = avpkt->size; float *buf_out = data; // pointer to the output data buffer int i, subframe; // counters - enum Mode speech_mode = MODE_475; // ??? + enum Mode speech_mode = MODE_475; // mode indication for DTX frames float fixed_gain_factor; float exc_feedback[AMR_SUBFRAME_SIZE]; float fixed_vector[AMR_SUBFRAME_SIZE]; // algebraic code book (fixed) vector

1 0

[soc]: r4849 - als/alsdec.c
by thilo.borgmann 31 Jul '09

31 Jul '09

Author: thilo.borgmann Date: Fri Jul 31 01:52:13 2009 New Revision: 4849 Log: Removed all-caps variable names. Modified: als/alsdec.c Modified: als/alsdec.c ============================================================================== --- als/alsdec.c Fri Jul 31 01:41:59 2009 (r4848) +++ als/alsdec.c Fri Jul 31 01:52:13 2009 (r4849) @@ -484,7 +484,7 @@ static int read_block_data(ALSDecContext quant_index = decode_rice(gb, rice_param) + offset; quant_cof[1] = -parcor_scaled_values[quant_index + 64]; - // read coefficients 2 - 20 + // read coefficients 2 - 19 k_max = FFMIN(20, opt_order); for (k = 2; k < k_max; k++) { offset = parcor_rice_table[sconf->coef_table][k][0]; @@ -493,8 +493,8 @@ static int read_block_data(ALSDecContext quant_cof[k] = (quant_index << 14) + (1 << 13); } - // read coefficients 20 - 128 - k_max = FFMIN(128, opt_order); + // read coefficients 20 - 126 + k_max = FFMIN(127, opt_order); for (k = 20; k < k_max; k++) { offset = k & 1; rice_param = 2; @@ -502,8 +502,8 @@ static int read_block_data(ALSDecContext quant_cof[k] = (quant_index << 14) + (1 << 13); } - // read coefficients 128 - opt_order - for (k = 128; k < opt_order; k++) { + // read coefficients 127 - opt_order + for (k = 127; k < opt_order; k++) { offset = 0; rice_param = 1; quant_index = decode_rice(gb, rice_param) + offset; @@ -675,21 +675,21 @@ static int read_frame_data(ALSDecContext raw_samples_R = ctx->raw_samples[c ] + sconf->frame_length; while (js_blocks[0] || js_blocks[1]) { - unsigned int LD, RD; + unsigned int diff_l, diff_r; block_length = sconf->frame_length >> div_blocks[b]; raw_samples_L -= block_length; raw_samples_R -= block_length; - LD = js_blocks[0] & 1; - RD = js_blocks[1] & 1; + diff_l = js_blocks[0] & 1; + diff_r = js_blocks[1] & 1; - if (LD) { // L = R - D - if (RD) + if (diff_l) { // L = R - D + if (diff_r) av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel pair!"); for (s = 0; s < block_length; s++) raw_samples_L[s] = raw_samples_R[s] - raw_samples_L[s]; - } else if (RD) { // R = D + L + } else if (diff_r) { // R = D + L for (s = 0; s < block_length; s++) raw_samples_R[s] = raw_samples_R[s] + raw_samples_L[s]; }

3 3

[soc]: r4858 - amr/amrnbdec.c
by cmcq 31 Jul '09

31 Jul '09

Author: cmcq Date: Fri Jul 31 06:43:12 2009 New Revision: 4858 Log: Make an internal function static Modified: amr/amrnbdec.c Modified: amr/amrnbdec.c ============================================================================== --- amr/amrnbdec.c Fri Jul 31 05:11:53 2009 (r4857) +++ amr/amrnbdec.c Fri Jul 31 06:43:12 2009 (r4858) @@ -209,7 +209,7 @@ static void interpolate_lsf(float lsf_q[ * Omitting this step creates audible distortion on the sinusoidal sweep * test vectors in 3GPP TS 26.074. */ -void adjust_lsfs(float *m) { +static void adjust_lsfs(float *m) { int i; float tmp=0.0; for (i = 0; i < LP_FILTER_ORDER; i++)

1 0

[soc]: r4857 - amr/amrnbdec.c
by cmcq 31 Jul '09

31 Jul '09

Author: cmcq Date: Fri Jul 31 05:11:53 2009 New Revision: 4857 Log: Remove section comments; the function names should be sufficiently clear Modified: amr/amrnbdec.c Modified: amr/amrnbdec.c ============================================================================== --- amr/amrnbdec.c Fri Jul 31 05:08:18 2009 (r4856) +++ amr/amrnbdec.c Fri Jul 31 05:11:53 2009 (r4857) @@ -1117,7 +1117,6 @@ static int amrnb_decode_frame(AVCodecCon av_log_missing_feature(avctx, "dtx mode", 1); return -1; } -/*** LPC coefficient decoding ***/ if (p->cur_frame_mode == MODE_122) { // decode split-matrix quantized lsf vector indexes to lsp vectors @@ -1131,14 +1130,10 @@ static int amrnb_decode_frame(AVCodecCon for (i = 0; i < 4; i++) lsp2lpc(p->lsp[i], p->lpc[i]); -/*** end of LPC coefficient decoding ***/ - for (subframe = 0; subframe < 4; subframe++) { const AMRNBSubframe *amr_subframe = &p->frame.subframe[subframe]; -/*** adaptive code book (pitch) vector decoding ***/ - decode_pitch_vector(p, amr_subframe, subframe); -/*** end of adaptive code book (pitch) vector decoding ***/ + decode_pitch_vector(p, amr_subframe, subframe); decode_fixed_vector(fixed_vector, amr_subframe->pulses, p->cur_frame_mode, subframe); @@ -1155,8 +1150,6 @@ static int amrnb_decode_frame(AVCodecCon set_fixed_gain(p, p->cur_frame_mode, fixed_gain_factor, fixed_vector); -/*** pre-processing ***/ - // The excitation feedback is calculated without any processing such // as fixed gain smoothing. This isn't mentioned in the specification. ff_weighted_vector_sumf(exc_feedback, p->excitation, fixed_vector, @@ -1172,10 +1165,6 @@ static int amrnb_decode_frame(AVCodecCon synth_fixed_vector = anti_sparseness(p, fixed_vector, synth_fixed_gain, spare_vector); -/*** end of pre-processing ***/ - -/*** synthesis ***/ - if (synthesis(p, p->excitation, p->lpc[subframe], synth_fixed_gain, synth_fixed_vector, &p->samples_in[LP_FILTER_ORDER], 0)) // overflow detected -> rerun synthesis scaling pitch vector down @@ -1184,8 +1173,6 @@ static int amrnb_decode_frame(AVCodecCon synthesis(p, p->excitation, p->lpc[subframe], synth_fixed_gain, synth_fixed_vector, &p->samples_in[LP_FILTER_ORDER], 1); -/*** end of synthesis ***/ - postfilter(p, p->lpc[subframe], buf_out + subframe * AMR_SUBFRAME_SIZE); // update buffers and history

1 0

[soc]: r4856 - amr/amrnbdec.c
by cmcq 31 Jul '09

31 Jul '09

Author: cmcq Date: Fri Jul 31 05:08:18 2009 New Revision: 4856 Log: Use multiply instead of divide, for speed Modified: amr/amrnbdec.c Modified: amr/amrnbdec.c ============================================================================== --- amr/amrnbdec.c Fri Jul 31 05:05:57 2009 (r4855) +++ amr/amrnbdec.c Fri Jul 31 05:08:18 2009 (r4856) @@ -916,7 +916,7 @@ static int synthesis(AMRContext *p, floa // factor of 4 if (overflow) for (i = 0; i < AMR_SUBFRAME_SIZE; i++) - p->pitch_vector[i] /= 4.0; + p->pitch_vector[i] *= 0.25; // construct the excitation vector for (i = 0; i < AMR_SUBFRAME_SIZE; i++)

1 0

[soc]: r4855 - amr/amrnbdec.c
by cmcq 31 Jul '09

31 Jul '09

Author: cmcq Date: Fri Jul 31 05:05:57 2009 New Revision: 4855 Log: Use av_clipf Modified: amr/amrnbdec.c Modified: amr/amrnbdec.c ============================================================================== --- amr/amrnbdec.c Fri Jul 31 05:02:37 2009 (r4854) +++ amr/amrnbdec.c Fri Jul 31 05:05:57 2009 (r4855) @@ -750,7 +750,7 @@ static float fixed_gain_smooth(AMRContex p->hang_count++; } else if (mode < MODE_74 || mode == MODE_102) { // calculate the fixed gain smoothing factor (k_m) - const float smoothing_factor = FFMIN(1.0, FFMAX(0.0, 4.0*diff - 1.6)); + const float smoothing_factor = av_clipf(4.0 * diff - 1.6, 0.0, 1.0); // calculate the mean fixed gain for the current subframe const float fixed_gain_mean = (p->fixed_gain[0] + p->fixed_gain[1] + p->fixed_gain[2] + p->fixed_gain[3] + p->fixed_gain[4])*0.2; // calculate the smoothed fixed gain

1 0

[soc]: r4854 - amr/amrnbdec.c
by cmcq 31 Jul '09

31 Jul '09

Author: cmcq Date: Fri Jul 31 05:02:37 2009 New Revision: 4854 Log: Use appropriate enum types Modified: amr/amrnbdec.c Modified: amr/amrnbdec.c ============================================================================== --- amr/amrnbdec.c Fri Jul 31 03:59:27 2009 (r4853) +++ amr/amrnbdec.c Fri Jul 31 05:02:37 2009 (r4854) @@ -47,8 +47,8 @@ typedef struct AMRContext { AMRNBFrame frame; ///< decoded AMR parameters (lsf coefficients, codebook indexes, etc) int bad_frame_indicator; ///< bad frame ? 1 : 0 - int cur_frame_mode; ///< current frame mode - int cur_frame_type; ///< current frame type + enum Mode cur_frame_mode; ///< current frame mode + enum RXFrameType cur_frame_type; ///< current frame type float prev_lsf_r[LP_FILTER_ORDER]; ///< residual LSF vector from previous subframe float lsp[4][LP_FILTER_ORDER]; ///< lsp vectors from current frame

1 0