yading@10: /*
yading@10:  * DSP Group TrueSpeech compatible decoder
yading@10:  * Copyright (c) 2005 Konstantin Shishkov
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include "libavutil/channel_layout.h"
yading@10: #include "libavutil/intreadwrite.h"
yading@10: #include "avcodec.h"
yading@10: #include "dsputil.h"
yading@10: #include "get_bits.h"
yading@10: #include "internal.h"
yading@10: 
yading@10: #include "truespeech_data.h"
yading@10: /**
yading@10:  * @file
yading@10:  * TrueSpeech decoder.
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * TrueSpeech decoder context
yading@10:  */
yading@10: typedef struct {
yading@10:     DSPContext dsp;
yading@10:     /* input data */
yading@10:     DECLARE_ALIGNED(16, uint8_t, buffer)[32];
yading@10:     int16_t vector[8];  ///< input vector: 5/5/4/4/4/3/3/3
yading@10:     int offset1[2];     ///< 8-bit value, used in one copying offset
yading@10:     int offset2[4];     ///< 7-bit value, encodes offsets for copying and for two-point filter
yading@10:     int pulseoff[4];    ///< 4-bit offset of pulse values block
yading@10:     int pulsepos[4];    ///< 27-bit variable, encodes 7 pulse positions
yading@10:     int pulseval[4];    ///< 7x2-bit pulse values
yading@10:     int flag;           ///< 1-bit flag, shows how to choose filters
yading@10:     /* temporary data */
yading@10:     int filtbuf[146];   // some big vector used for storing filters
yading@10:     int prevfilt[8];    // filter from previous frame
yading@10:     int16_t tmp1[8];    // coefficients for adding to out
yading@10:     int16_t tmp2[8];    // coefficients for adding to out
yading@10:     int16_t tmp3[8];    // coefficients for adding to out
yading@10:     int16_t cvector[8]; // correlated input vector
yading@10:     int filtval;        // gain value for one function
yading@10:     int16_t newvec[60]; // tmp vector
yading@10:     int16_t filters[32]; // filters for every subframe
yading@10: } TSContext;
yading@10: 
yading@10: static av_cold int truespeech_decode_init(AVCodecContext * avctx)
yading@10: {
yading@10:     TSContext *c = avctx->priv_data;
yading@10: 
yading@10:     if (avctx->channels != 1) {
yading@10:         avpriv_request_sample(avctx, "Channel count %d", avctx->channels);
yading@10:         return AVERROR_PATCHWELCOME;
yading@10:     }
yading@10: 
yading@10:     avctx->channel_layout = AV_CH_LAYOUT_MONO;
yading@10:     avctx->sample_fmt     = AV_SAMPLE_FMT_S16;
yading@10: 
yading@10:     ff_dsputil_init(&c->dsp, avctx);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
yading@10: {
yading@10:     GetBitContext gb;
yading@10: 
yading@10:     dec->dsp.bswap_buf((uint32_t *)dec->buffer, (const uint32_t *)input, 8);
yading@10:     init_get_bits(&gb, dec->buffer, 32 * 8);
yading@10: 
yading@10:     dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)];
yading@10:     dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)];
yading@10:     dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)];
yading@10:     dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)];
yading@10:     dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)];
yading@10:     dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)];
yading@10:     dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)];
yading@10:     dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)];
yading@10:     dec->flag      = get_bits1(&gb);
yading@10: 
yading@10:     dec->offset1[0] = get_bits(&gb, 4) << 4;
yading@10:     dec->offset2[3] = get_bits(&gb, 7);
yading@10:     dec->offset2[2] = get_bits(&gb, 7);
yading@10:     dec->offset2[1] = get_bits(&gb, 7);
yading@10:     dec->offset2[0] = get_bits(&gb, 7);
yading@10: 
yading@10:     dec->offset1[1]  = get_bits(&gb, 4);
yading@10:     dec->pulseval[1] = get_bits(&gb, 14);
yading@10:     dec->pulseval[0] = get_bits(&gb, 14);
yading@10: 
yading@10:     dec->offset1[1] |= get_bits(&gb, 4) << 4;
yading@10:     dec->pulseval[3] = get_bits(&gb, 14);
yading@10:     dec->pulseval[2] = get_bits(&gb, 14);
yading@10: 
yading@10:     dec->offset1[0] |= get_bits1(&gb);
yading@10:     dec->pulsepos[0] = get_bits_long(&gb, 27);
yading@10:     dec->pulseoff[0] = get_bits(&gb, 4);
yading@10: 
yading@10:     dec->offset1[0] |= get_bits1(&gb) << 1;
yading@10:     dec->pulsepos[1] = get_bits_long(&gb, 27);
yading@10:     dec->pulseoff[1] = get_bits(&gb, 4);
yading@10: 
yading@10:     dec->offset1[0] |= get_bits1(&gb) << 2;
yading@10:     dec->pulsepos[2] = get_bits_long(&gb, 27);
yading@10:     dec->pulseoff[2] = get_bits(&gb, 4);
yading@10: 
yading@10:     dec->offset1[0] |= get_bits1(&gb) << 3;
yading@10:     dec->pulsepos[3] = get_bits_long(&gb, 27);
yading@10:     dec->pulseoff[3] = get_bits(&gb, 4);
yading@10: }
yading@10: 
yading@10: static void truespeech_correlate_filter(TSContext *dec)
yading@10: {
yading@10:     int16_t tmp[8];
yading@10:     int i, j;
yading@10: 
yading@10:     for(i = 0; i < 8; i++){
yading@10:         if(i > 0){
yading@10:             memcpy(tmp, dec->cvector, i * sizeof(*tmp));
yading@10:             for(j = 0; j < i; j++)
yading@10:                 dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) +
yading@10:                                    (dec->cvector[j] << 15) + 0x4000) >> 15;
yading@10:         }
yading@10:         dec->cvector[i] = (8 - dec->vector[i]) >> 3;
yading@10:     }
yading@10:     for(i = 0; i < 8; i++)
yading@10:         dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15;
yading@10: 
yading@10:     dec->filtval = dec->vector[0];
yading@10: }
yading@10: 
yading@10: static void truespeech_filters_merge(TSContext *dec)
yading@10: {
yading@10:     int i;
yading@10: 
yading@10:     if(!dec->flag){
yading@10:         for(i = 0; i < 8; i++){
yading@10:             dec->filters[i + 0] = dec->prevfilt[i];
yading@10:             dec->filters[i + 8] = dec->prevfilt[i];
yading@10:         }
yading@10:     }else{
yading@10:         for(i = 0; i < 8; i++){
yading@10:             dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
yading@10:             dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
yading@10:         }
yading@10:     }
yading@10:     for(i = 0; i < 8; i++){
yading@10:         dec->filters[i + 16] = dec->cvector[i];
yading@10:         dec->filters[i + 24] = dec->cvector[i];
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
yading@10: {
yading@10:     int16_t tmp[146 + 60], *ptr0, *ptr1;
yading@10:     const int16_t *filter;
yading@10:     int i, t, off;
yading@10: 
yading@10:     t = dec->offset2[quart];
yading@10:     if(t == 127){
yading@10:         memset(dec->newvec, 0, 60 * sizeof(*dec->newvec));
yading@10:         return;
yading@10:     }
yading@10:     for(i = 0; i < 146; i++)
yading@10:         tmp[i] = dec->filtbuf[i];
yading@10:     off = (t / 25) + dec->offset1[quart >> 1] + 18;
yading@10:     off = av_clip(off, 0, 145);
yading@10:     ptr0 = tmp + 145 - off;
yading@10:     ptr1 = tmp + 146;
yading@10:     filter = ts_order2_coeffs + (t % 25) * 2;
yading@10:     for(i = 0; i < 60; i++){
yading@10:         t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
yading@10:         ptr0++;
yading@10:         dec->newvec[i] = t;
yading@10:         ptr1[i] = t;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
yading@10: {
yading@10:     int16_t tmp[7];
yading@10:     int i, j, t;
yading@10:     const int16_t *ptr1;
yading@10:     int16_t *ptr2;
yading@10:     int coef;
yading@10: 
yading@10:     memset(out, 0, 60 * sizeof(*out));
yading@10:     for(i = 0; i < 7; i++) {
yading@10:         t = dec->pulseval[quart] & 3;
yading@10:         dec->pulseval[quart] >>= 2;
yading@10:         tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t];
yading@10:     }
yading@10: 
yading@10:     coef = dec->pulsepos[quart] >> 15;
yading@10:     ptr1 = ts_pulse_values + 30;
yading@10:     ptr2 = tmp;
yading@10:     for(i = 0, j = 3; (i < 30) && (j > 0); i++){
yading@10:         t = *ptr1++;
yading@10:         if(coef >= t)
yading@10:             coef -= t;
yading@10:         else{
yading@10:             out[i] = *ptr2++;
yading@10:             ptr1 += 30;
yading@10:             j--;
yading@10:         }
yading@10:     }
yading@10:     coef = dec->pulsepos[quart] & 0x7FFF;
yading@10:     ptr1 = ts_pulse_values;
yading@10:     for(i = 30, j = 4; (i < 60) && (j > 0); i++){
yading@10:         t = *ptr1++;
yading@10:         if(coef >= t)
yading@10:             coef -= t;
yading@10:         else{
yading@10:             out[i] = *ptr2++;
yading@10:             ptr1 += 30;
yading@10:             j--;
yading@10:         }
yading@10:     }
yading@10: 
yading@10: }
yading@10: 
yading@10: static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
yading@10: {
yading@10:     int i;
yading@10: 
yading@10:     memmove(dec->filtbuf, &dec->filtbuf[60], 86 * sizeof(*dec->filtbuf));
yading@10:     for(i = 0; i < 60; i++){
yading@10:         dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
yading@10:         out[i] += dec->newvec[i];
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
yading@10: {
yading@10:     int i,k;
yading@10:     int t[8];
yading@10:     int16_t *ptr0, *ptr1;
yading@10: 
yading@10:     ptr0 = dec->tmp1;
yading@10:     ptr1 = dec->filters + quart * 8;
yading@10:     for(i = 0; i < 60; i++){
yading@10:         int sum = 0;
yading@10:         for(k = 0; k < 8; k++)
yading@10:             sum += ptr0[k] * ptr1[k];
yading@10:         sum = (sum + (out[i] << 12) + 0x800) >> 12;
yading@10:         out[i] = av_clip(sum, -0x7FFE, 0x7FFE);
yading@10:         for(k = 7; k > 0; k--)
yading@10:             ptr0[k] = ptr0[k - 1];
yading@10:         ptr0[0] = out[i];
yading@10:     }
yading@10: 
yading@10:     for(i = 0; i < 8; i++)
yading@10:         t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15;
yading@10: 
yading@10:     ptr0 = dec->tmp2;
yading@10:     for(i = 0; i < 60; i++){
yading@10:         int sum = 0;
yading@10:         for(k = 0; k < 8; k++)
yading@10:             sum += ptr0[k] * t[k];
yading@10:         for(k = 7; k > 0; k--)
yading@10:             ptr0[k] = ptr0[k - 1];
yading@10:         ptr0[0] = out[i];
yading@10:         out[i] = ((out[i] << 12) - sum) >> 12;
yading@10:     }
yading@10: 
yading@10:     for(i = 0; i < 8; i++)
yading@10:         t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15;
yading@10: 
yading@10:     ptr0 = dec->tmp3;
yading@10:     for(i = 0; i < 60; i++){
yading@10:         int sum = out[i] << 12;
yading@10:         for(k = 0; k < 8; k++)
yading@10:             sum += ptr0[k] * t[k];
yading@10:         for(k = 7; k > 0; k--)
yading@10:             ptr0[k] = ptr0[k - 1];
yading@10:         ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
yading@10: 
yading@10:         sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
yading@10:         sum = sum - (sum >> 3);
yading@10:         out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void truespeech_save_prevvec(TSContext *c)
yading@10: {
yading@10:     int i;
yading@10: 
yading@10:     for(i = 0; i < 8; i++)
yading@10:         c->prevfilt[i] = c->cvector[i];
yading@10: }
yading@10: 
yading@10: static int truespeech_decode_frame(AVCodecContext *avctx, void *data,
yading@10:                                    int *got_frame_ptr, AVPacket *avpkt)
yading@10: {
yading@10:     AVFrame *frame     = data;
yading@10:     const uint8_t *buf = avpkt->data;
yading@10:     int buf_size = avpkt->size;
yading@10:     TSContext *c = avctx->priv_data;
yading@10: 
yading@10:     int i, j;
yading@10:     int16_t *samples;
yading@10:     int iterations, ret;
yading@10: 
yading@10:     iterations = buf_size / 32;
yading@10: 
yading@10:     if (!iterations) {
yading@10:         av_log(avctx, AV_LOG_ERROR,
yading@10:                "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
yading@10:         return -1;
yading@10:     }
yading@10: 
yading@10:     /* get output buffer */
yading@10:     frame->nb_samples = iterations * 240;
yading@10:     if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
yading@10:         return ret;
yading@10:     samples = (int16_t *)frame->data[0];
yading@10: 
yading@10:     memset(samples, 0, iterations * 240 * sizeof(*samples));
yading@10: 
yading@10:     for(j = 0; j < iterations; j++) {
yading@10:         truespeech_read_frame(c, buf);
yading@10:         buf += 32;
yading@10: 
yading@10:         truespeech_correlate_filter(c);
yading@10:         truespeech_filters_merge(c);
yading@10: 
yading@10:         for(i = 0; i < 4; i++) {
yading@10:             truespeech_apply_twopoint_filter(c, i);
yading@10:             truespeech_place_pulses  (c, samples, i);
yading@10:             truespeech_update_filters(c, samples, i);
yading@10:             truespeech_synth         (c, samples, i);
yading@10:             samples += 60;
yading@10:         }
yading@10: 
yading@10:         truespeech_save_prevvec(c);
yading@10:     }
yading@10: 
yading@10:     *got_frame_ptr = 1;
yading@10: 
yading@10:     return buf_size;
yading@10: }
yading@10: 
yading@10: AVCodec ff_truespeech_decoder = {
yading@10:     .name           = "truespeech",
yading@10:     .type           = AVMEDIA_TYPE_AUDIO,
yading@10:     .id             = AV_CODEC_ID_TRUESPEECH,
yading@10:     .priv_data_size = sizeof(TSContext),
yading@10:     .init           = truespeech_decode_init,
yading@10:     .decode         = truespeech_decode_frame,
yading@10:     .capabilities   = CODEC_CAP_DR1,
yading@10:     .long_name      = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),
yading@10: };