yading@11: /*
yading@11:  * Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
yading@11:  *
yading@11:  * Triangular with Noise Shaping is based on opusfile.
yading@11:  * Copyright (c) 1994-2012 by the Xiph.Org Foundation and contributors
yading@11:  *
yading@11:  * This file is part of Libav.
yading@11:  *
yading@11:  * Libav is free software; you can redistribute it and/or
yading@11:  * modify it under the terms of the GNU Lesser General Public
yading@11:  * License as published by the Free Software Foundation; either
yading@11:  * version 2.1 of the License, or (at your option) any later version.
yading@11:  *
yading@11:  * Libav is distributed in the hope that it will be useful,
yading@11:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@11:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@11:  * Lesser General Public License for more details.
yading@11:  *
yading@11:  * You should have received a copy of the GNU Lesser General Public
yading@11:  * License along with Libav; if not, write to the Free Software
yading@11:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@11:  */
yading@11: 
yading@11: /**
yading@11:  * @file
yading@11:  * Dithered Audio Sample Quantization
yading@11:  *
yading@11:  * Converts from dbl, flt, or s32 to s16 using dithering.
yading@11:  */
yading@11: 
yading@11: #include <math.h>
yading@11: #include <stdint.h>
yading@11: 
yading@11: #include "libavutil/common.h"
yading@11: #include "libavutil/lfg.h"
yading@11: #include "libavutil/mem.h"
yading@11: #include "libavutil/samplefmt.h"
yading@11: #include "audio_convert.h"
yading@11: #include "dither.h"
yading@11: #include "internal.h"
yading@11: 
yading@11: typedef struct DitherState {
yading@11:     int mute;
yading@11:     unsigned int seed;
yading@11:     AVLFG lfg;
yading@11:     float *noise_buf;
yading@11:     int noise_buf_size;
yading@11:     int noise_buf_ptr;
yading@11:     float dither_a[4];
yading@11:     float dither_b[4];
yading@11: } DitherState;
yading@11: 
yading@11: struct DitherContext {
yading@11:     DitherDSPContext  ddsp;
yading@11:     enum AVResampleDitherMethod method;
yading@11:     int apply_map;
yading@11:     ChannelMapInfo *ch_map_info;
yading@11: 
yading@11:     int mute_dither_threshold;  // threshold for disabling dither
yading@11:     int mute_reset_threshold;   // threshold for resetting noise shaping
yading@11:     const float *ns_coef_b;     // noise shaping coeffs
yading@11:     const float *ns_coef_a;     // noise shaping coeffs
yading@11: 
yading@11:     int channels;
yading@11:     DitherState *state;         // dither states for each channel
yading@11: 
yading@11:     AudioData *flt_data;        // input data in fltp
yading@11:     AudioData *s16_data;        // dithered output in s16p
yading@11:     AudioConvert *ac_in;        // converter for input to fltp
yading@11:     AudioConvert *ac_out;       // converter for s16p to s16 (if needed)
yading@11: 
yading@11:     void (*quantize)(int16_t *dst, const float *src, float *dither, int len);
yading@11:     int samples_align;
yading@11: };
yading@11: 
yading@11: /* mute threshold, in seconds */
yading@11: #define MUTE_THRESHOLD_SEC 0.000333
yading@11: 
yading@11: /* scale factor for 16-bit output.
yading@11:    The signal is attenuated slightly to avoid clipping */
yading@11: #define S16_SCALE 32753.0f
yading@11: 
yading@11: /* scale to convert lfg from INT_MIN/INT_MAX to -0.5/0.5 */
yading@11: #define LFG_SCALE (1.0f / (2.0f * INT32_MAX))
yading@11: 
yading@11: /* noise shaping coefficients */
yading@11: 
yading@11: static const float ns_48_coef_b[4] = {
yading@11:     2.2374f, -0.7339f, -0.1251f, -0.6033f
yading@11: };
yading@11: 
yading@11: static const float ns_48_coef_a[4] = {
yading@11:     0.9030f, 0.0116f, -0.5853f, -0.2571f
yading@11: };
yading@11: 
yading@11: static const float ns_44_coef_b[4] = {
yading@11:     2.2061f, -0.4707f, -0.2534f, -0.6213f
yading@11: };
yading@11: 
yading@11: static const float ns_44_coef_a[4] = {
yading@11:     1.0587f, 0.0676f, -0.6054f, -0.2738f
yading@11: };
yading@11: 
yading@11: static void dither_int_to_float_rectangular_c(float *dst, int *src, int len)
yading@11: {
yading@11:     int i;
yading@11:     for (i = 0; i < len; i++)
yading@11:         dst[i] = src[i] * LFG_SCALE;
yading@11: }
yading@11: 
yading@11: static void dither_int_to_float_triangular_c(float *dst, int *src0, int len)
yading@11: {
yading@11:     int i;
yading@11:     int *src1  = src0 + len;
yading@11: 
yading@11:     for (i = 0; i < len; i++) {
yading@11:         float r = src0[i] * LFG_SCALE;
yading@11:         r      += src1[i] * LFG_SCALE;
yading@11:         dst[i]  = r;
yading@11:     }
yading@11: }
yading@11: 
yading@11: static void quantize_c(int16_t *dst, const float *src, float *dither, int len)
yading@11: {
yading@11:     int i;
yading@11:     for (i = 0; i < len; i++)
yading@11:         dst[i] = av_clip_int16(lrintf(src[i] * S16_SCALE + dither[i]));
yading@11: }
yading@11: 
yading@11: #define SQRT_1_6 0.40824829046386301723f
yading@11: 
yading@11: static void dither_highpass_filter(float *src, int len)
yading@11: {
yading@11:     int i;
yading@11: 
yading@11:     /* filter is from libswresample in FFmpeg */
yading@11:     for (i = 0; i < len - 2; i++)
yading@11:         src[i] = (-src[i] + 2 * src[i + 1] - src[i + 2]) * SQRT_1_6;
yading@11: }
yading@11: 
yading@11: static int generate_dither_noise(DitherContext *c, DitherState *state,
yading@11:                                  int min_samples)
yading@11: {
yading@11:     int i;
yading@11:     int nb_samples  = FFALIGN(min_samples, 16) + 16;
yading@11:     int buf_samples = nb_samples *
yading@11:                       (c->method == AV_RESAMPLE_DITHER_RECTANGULAR ? 1 : 2);
yading@11:     unsigned int *noise_buf_ui;
yading@11: 
yading@11:     av_freep(&state->noise_buf);
yading@11:     state->noise_buf_size = state->noise_buf_ptr = 0;
yading@11: 
yading@11:     state->noise_buf = av_malloc(buf_samples * sizeof(*state->noise_buf));
yading@11:     if (!state->noise_buf)
yading@11:         return AVERROR(ENOMEM);
yading@11:     state->noise_buf_size = FFALIGN(min_samples, 16);
yading@11:     noise_buf_ui          = (unsigned int *)state->noise_buf;
yading@11: 
yading@11:     av_lfg_init(&state->lfg, state->seed);
yading@11:     for (i = 0; i < buf_samples; i++)
yading@11:         noise_buf_ui[i] = av_lfg_get(&state->lfg);
yading@11: 
yading@11:     c->ddsp.dither_int_to_float(state->noise_buf, noise_buf_ui, nb_samples);
yading@11: 
yading@11:     if (c->method == AV_RESAMPLE_DITHER_TRIANGULAR_HP)
yading@11:         dither_highpass_filter(state->noise_buf, nb_samples);
yading@11: 
yading@11:     return 0;
yading@11: }
yading@11: 
yading@11: static void quantize_triangular_ns(DitherContext *c, DitherState *state,
yading@11:                                    int16_t *dst, const float *src,
yading@11:                                    int nb_samples)
yading@11: {
yading@11:     int i, j;
yading@11:     float *dither = &state->noise_buf[state->noise_buf_ptr];
yading@11: 
yading@11:     if (state->mute > c->mute_reset_threshold)
yading@11:         memset(state->dither_a, 0, sizeof(state->dither_a));
yading@11: 
yading@11:     for (i = 0; i < nb_samples; i++) {
yading@11:         float err = 0;
yading@11:         float sample = src[i] * S16_SCALE;
yading@11: 
yading@11:         for (j = 0; j < 4; j++) {
yading@11:             err += c->ns_coef_b[j] * state->dither_b[j] -
yading@11:                    c->ns_coef_a[j] * state->dither_a[j];
yading@11:         }
yading@11:         for (j = 3; j > 0; j--) {
yading@11:             state->dither_a[j] = state->dither_a[j - 1];
yading@11:             state->dither_b[j] = state->dither_b[j - 1];
yading@11:         }
yading@11:         state->dither_a[0] = err;
yading@11:         sample -= err;
yading@11: 
yading@11:         if (state->mute > c->mute_dither_threshold) {
yading@11:             dst[i]             = av_clip_int16(lrintf(sample));
yading@11:             state->dither_b[0] = 0;
yading@11:         } else {
yading@11:             dst[i]             = av_clip_int16(lrintf(sample + dither[i]));
yading@11:             state->dither_b[0] = av_clipf(dst[i] - sample, -1.5f, 1.5f);
yading@11:         }
yading@11: 
yading@11:         state->mute++;
yading@11:         if (src[i])
yading@11:             state->mute = 0;
yading@11:     }
yading@11: }
yading@11: 
yading@11: static int convert_samples(DitherContext *c, int16_t **dst, float * const *src,
yading@11:                            int channels, int nb_samples)
yading@11: {
yading@11:     int ch, ret;
yading@11:     int aligned_samples = FFALIGN(nb_samples, 16);
yading@11: 
yading@11:     for (ch = 0; ch < channels; ch++) {
yading@11:         DitherState *state = &c->state[ch];
yading@11: 
yading@11:         if (state->noise_buf_size < aligned_samples) {
yading@11:             ret = generate_dither_noise(c, state, nb_samples);
yading@11:             if (ret < 0)
yading@11:                 return ret;
yading@11:         } else if (state->noise_buf_size - state->noise_buf_ptr < aligned_samples) {
yading@11:             state->noise_buf_ptr = 0;
yading@11:         }
yading@11: 
yading@11:         if (c->method == AV_RESAMPLE_DITHER_TRIANGULAR_NS) {
yading@11:             quantize_triangular_ns(c, state, dst[ch], src[ch], nb_samples);
yading@11:         } else {
yading@11:             c->quantize(dst[ch], src[ch],
yading@11:                         &state->noise_buf[state->noise_buf_ptr],
yading@11:                         FFALIGN(nb_samples, c->samples_align));
yading@11:         }
yading@11: 
yading@11:         state->noise_buf_ptr += aligned_samples;
yading@11:     }
yading@11: 
yading@11:     return 0;
yading@11: }
yading@11: 
yading@11: int ff_convert_dither(DitherContext *c, AudioData *dst, AudioData *src)
yading@11: {
yading@11:     int ret;
yading@11:     AudioData *flt_data;
yading@11: 
yading@11:     /* output directly to dst if it is planar */
yading@11:     if (dst->sample_fmt == AV_SAMPLE_FMT_S16P)
yading@11:         c->s16_data = dst;
yading@11:     else {
yading@11:         /* make sure s16_data is large enough for the output */
yading@11:         ret = ff_audio_data_realloc(c->s16_data, src->nb_samples);
yading@11:         if (ret < 0)
yading@11:             return ret;
yading@11:     }
yading@11: 
yading@11:     if (src->sample_fmt != AV_SAMPLE_FMT_FLTP || c->apply_map) {
yading@11:         /* make sure flt_data is large enough for the input */
yading@11:         ret = ff_audio_data_realloc(c->flt_data, src->nb_samples);
yading@11:         if (ret < 0)
yading@11:             return ret;
yading@11:         flt_data = c->flt_data;
yading@11:     }
yading@11: 
yading@11:     if (src->sample_fmt != AV_SAMPLE_FMT_FLTP) {
yading@11:         /* convert input samples to fltp and scale to s16 range */
yading@11:         ret = ff_audio_convert(c->ac_in, flt_data, src);
yading@11:         if (ret < 0)
yading@11:             return ret;
yading@11:     } else if (c->apply_map) {
yading@11:         ret = ff_audio_data_copy(flt_data, src, c->ch_map_info);
yading@11:         if (ret < 0)
yading@11:             return ret;
yading@11:     } else {
yading@11:         flt_data = src;
yading@11:     }
yading@11: 
yading@11:     /* check alignment and padding constraints */
yading@11:     if (c->method != AV_RESAMPLE_DITHER_TRIANGULAR_NS) {
yading@11:         int ptr_align     = FFMIN(flt_data->ptr_align,     c->s16_data->ptr_align);
yading@11:         int samples_align = FFMIN(flt_data->samples_align, c->s16_data->samples_align);
yading@11:         int aligned_len   = FFALIGN(src->nb_samples, c->ddsp.samples_align);
yading@11: 
yading@11:         if (!(ptr_align % c->ddsp.ptr_align) && samples_align >= aligned_len) {
yading@11:             c->quantize      = c->ddsp.quantize;
yading@11:             c->samples_align = c->ddsp.samples_align;
yading@11:         } else {
yading@11:             c->quantize      = quantize_c;
yading@11:             c->samples_align = 1;
yading@11:         }
yading@11:     }
yading@11: 
yading@11:     ret = convert_samples(c, (int16_t **)c->s16_data->data,
yading@11:                           (float * const *)flt_data->data, src->channels,
yading@11:                           src->nb_samples);
yading@11:     if (ret < 0)
yading@11:         return ret;
yading@11: 
yading@11:     c->s16_data->nb_samples = src->nb_samples;
yading@11: 
yading@11:     /* interleave output to dst if needed */
yading@11:     if (dst->sample_fmt == AV_SAMPLE_FMT_S16) {
yading@11:         ret = ff_audio_convert(c->ac_out, dst, c->s16_data);
yading@11:         if (ret < 0)
yading@11:             return ret;
yading@11:     } else
yading@11:         c->s16_data = NULL;
yading@11: 
yading@11:     return 0;
yading@11: }
yading@11: 
yading@11: void ff_dither_free(DitherContext **cp)
yading@11: {
yading@11:     DitherContext *c = *cp;
yading@11:     int ch;
yading@11: 
yading@11:     if (!c)
yading@11:         return;
yading@11:     ff_audio_data_free(&c->flt_data);
yading@11:     ff_audio_data_free(&c->s16_data);
yading@11:     ff_audio_convert_free(&c->ac_in);
yading@11:     ff_audio_convert_free(&c->ac_out);
yading@11:     for (ch = 0; ch < c->channels; ch++)
yading@11:         av_free(c->state[ch].noise_buf);
yading@11:     av_free(c->state);
yading@11:     av_freep(cp);
yading@11: }
yading@11: 
yading@11: static void dither_init(DitherDSPContext *ddsp,
yading@11:                         enum AVResampleDitherMethod method)
yading@11: {
yading@11:     ddsp->quantize      = quantize_c;
yading@11:     ddsp->ptr_align     = 1;
yading@11:     ddsp->samples_align = 1;
yading@11: 
yading@11:     if (method == AV_RESAMPLE_DITHER_RECTANGULAR)
yading@11:         ddsp->dither_int_to_float = dither_int_to_float_rectangular_c;
yading@11:     else
yading@11:         ddsp->dither_int_to_float = dither_int_to_float_triangular_c;
yading@11: 
yading@11:     if (ARCH_X86)
yading@11:         ff_dither_init_x86(ddsp, method);
yading@11: }
yading@11: 
yading@11: DitherContext *ff_dither_alloc(AVAudioResampleContext *avr,
yading@11:                                enum AVSampleFormat out_fmt,
yading@11:                                enum AVSampleFormat in_fmt,
yading@11:                                int channels, int sample_rate, int apply_map)
yading@11: {
yading@11:     AVLFG seed_gen;
yading@11:     DitherContext *c;
yading@11:     int ch;
yading@11: 
yading@11:     if (av_get_packed_sample_fmt(out_fmt) != AV_SAMPLE_FMT_S16 ||
yading@11:         av_get_bytes_per_sample(in_fmt) <= 2) {
yading@11:         av_log(avr, AV_LOG_ERROR, "dithering %s to %s is not supported\n",
yading@11:                av_get_sample_fmt_name(in_fmt), av_get_sample_fmt_name(out_fmt));
yading@11:         return NULL;
yading@11:     }
yading@11: 
yading@11:     c = av_mallocz(sizeof(*c));
yading@11:     if (!c)
yading@11:         return NULL;
yading@11: 
yading@11:     c->apply_map = apply_map;
yading@11:     if (apply_map)
yading@11:         c->ch_map_info = &avr->ch_map_info;
yading@11: 
yading@11:     if (avr->dither_method == AV_RESAMPLE_DITHER_TRIANGULAR_NS &&
yading@11:         sample_rate != 48000 && sample_rate != 44100) {
yading@11:         av_log(avr, AV_LOG_WARNING, "sample rate must be 48000 or 44100 Hz "
yading@11:                "for triangular_ns dither. using triangular_hp instead.\n");
yading@11:         avr->dither_method = AV_RESAMPLE_DITHER_TRIANGULAR_HP;
yading@11:     }
yading@11:     c->method = avr->dither_method;
yading@11:     dither_init(&c->ddsp, c->method);
yading@11: 
yading@11:     if (c->method == AV_RESAMPLE_DITHER_TRIANGULAR_NS) {
yading@11:         if (sample_rate == 48000) {
yading@11:             c->ns_coef_b = ns_48_coef_b;
yading@11:             c->ns_coef_a = ns_48_coef_a;
yading@11:         } else {
yading@11:             c->ns_coef_b = ns_44_coef_b;
yading@11:             c->ns_coef_a = ns_44_coef_a;
yading@11:         }
yading@11:     }
yading@11: 
yading@11:     /* Either s16 or s16p output format is allowed, but s16p is used
yading@11:        internally, so we need to use a temp buffer and interleave if the output
yading@11:        format is s16 */
yading@11:     if (out_fmt != AV_SAMPLE_FMT_S16P) {
yading@11:         c->s16_data = ff_audio_data_alloc(channels, 1024, AV_SAMPLE_FMT_S16P,
yading@11:                                           "dither s16 buffer");
yading@11:         if (!c->s16_data)
yading@11:             goto fail;
yading@11: 
yading@11:         c->ac_out = ff_audio_convert_alloc(avr, out_fmt, AV_SAMPLE_FMT_S16P,
yading@11:                                            channels, sample_rate, 0);
yading@11:         if (!c->ac_out)
yading@11:             goto fail;
yading@11:     }
yading@11: 
yading@11:     if (in_fmt != AV_SAMPLE_FMT_FLTP || c->apply_map) {
yading@11:         c->flt_data = ff_audio_data_alloc(channels, 1024, AV_SAMPLE_FMT_FLTP,
yading@11:                                           "dither flt buffer");
yading@11:         if (!c->flt_data)
yading@11:             goto fail;
yading@11:     }
yading@11:     if (in_fmt != AV_SAMPLE_FMT_FLTP) {
yading@11:         c->ac_in = ff_audio_convert_alloc(avr, AV_SAMPLE_FMT_FLTP, in_fmt,
yading@11:                                           channels, sample_rate, c->apply_map);
yading@11:         if (!c->ac_in)
yading@11:             goto fail;
yading@11:     }
yading@11: 
yading@11:     c->state = av_mallocz(channels * sizeof(*c->state));
yading@11:     if (!c->state)
yading@11:         goto fail;
yading@11:     c->channels = channels;
yading@11: 
yading@11:     /* calculate thresholds for turning off dithering during periods of
yading@11:        silence to avoid replacing digital silence with quiet dither noise */
yading@11:     c->mute_dither_threshold = lrintf(sample_rate * MUTE_THRESHOLD_SEC);
yading@11:     c->mute_reset_threshold  = c->mute_dither_threshold * 4;
yading@11: 
yading@11:     /* initialize dither states */
yading@11:     av_lfg_init(&seed_gen, 0xC0FFEE);
yading@11:     for (ch = 0; ch < channels; ch++) {
yading@11:         DitherState *state = &c->state[ch];
yading@11:         state->mute = c->mute_reset_threshold + 1;
yading@11:         state->seed = av_lfg_get(&seed_gen);
yading@11:         generate_dither_noise(c, state, FFMAX(32768, sample_rate / 2));
yading@11:     }
yading@11: 
yading@11:     return c;
yading@11: 
yading@11: fail:
yading@11:     ff_dither_free(&c);
yading@11:     return NULL;
yading@11: }