amrwbdec.c File Reference

#include "config.h"
#include "libavutil/channel_layout.h"
#include "libavutil/common.h"
#include "libavutil/lfg.h"
#include "avcodec.h"
#include "lsp.h"
#include "celp_filters.h"
#include "celp_math.h"
#include "acelp_filters.h"
#include "acelp_vectors.h"
#include "acelp_pitch_delay.h"
#include "codec_internal.h"
#include "decode.h"
#include "amr.h"
#include "amrwbdata.h"

Go to the source code of this file.

Data Structures
struct	AMRWBContext
struct	AMRWBChannelsContext

Macros
#define	AMR_USE_16BIT_TABLES
#define	BIT_STR(x, lsb, len)   av_mod_uintp2((x) >> (lsb), (len))
	Get x bits in the index interval [lsb,lsb+len-1] inclusive. More...
#define	BIT_POS(x, p)   (((x) >> (p)) & 1)
	Get the bit at specified position. More...

Functions
static av_cold int	amrwb_decode_init (AVCodecContext *avctx)
static int	decode_mime_header (AMRWBContext *ctx, const uint8_t *buf)
	Decode the frame header in the "MIME/storage" format. More...
static void	decode_isf_indices_36b (uint16_t *ind, float *isf_q)
	Decode quantized ISF vectors using 36-bit indexes (6K60 mode only). More...
static void	decode_isf_indices_46b (uint16_t *ind, float *isf_q)
	Decode quantized ISF vectors using 46-bit indexes (except 6K60 mode). More...
static void	isf_add_mean_and_past (float *isf_q, float *isf_past)
	Apply mean and past ISF values using the prediction factor. More...
static void	interpolate_isp (double isp_q[4][LP_ORDER], const double *isp4_past)
	Interpolate the fourth ISP vector from current and past frames to obtain an ISP vector for each subframe. More...
static void	decode_pitch_lag_high (int *lag_int, int *lag_frac, int pitch_index, uint8_t *base_lag_int, int subframe)
	Decode an adaptive codebook index into pitch lag (except 6k60, 8k85 modes). More...
static void	decode_pitch_lag_low (int *lag_int, int *lag_frac, int pitch_index, uint8_t *base_lag_int, int subframe, enum Mode mode)
	Decode an adaptive codebook index into pitch lag for 8k85 and 6k60 modes. More...
static void	decode_pitch_vector (AMRWBContext *ctx, const AMRWBSubFrame *amr_subframe, const int subframe)
	Find the pitch vector by interpolating the past excitation at the pitch delay, which is obtained in this function. More...
static void	decode_1p_track (int *out, int code, int m, int off)
	The next six functions decode_[i]p_track decode exactly i pulses positions and amplitudes (-1 or 1) in a subframe track using an encoded pulse indexing (TS 26.190 section 5.8.2). More...
static void	decode_2p_track (int *out, int code, int m, int off)
	code: 2m+1 bits More...
static void	decode_3p_track (int *out, int code, int m, int off)
	code: 3m+1 bits More...
static void	decode_4p_track (int *out, int code, int m, int off)
	code: 4m bits More...
static void	decode_5p_track (int *out, int code, int m, int off)
	code: 5m bits More...
static void	decode_6p_track (int *out, int code, int m, int off)
	code: 6m-2 bits More...
static void	decode_fixed_vector (float *fixed_vector, const uint16_t *pulse_hi, const uint16_t *pulse_lo, const enum Mode mode)
	Decode the algebraic codebook index to pulse positions and signs, then construct the algebraic codebook vector. More...
static void	decode_gains (const uint8_t vq_gain, const enum Mode mode, float *fixed_gain_factor, float *pitch_gain)
	Decode pitch gain and fixed gain correction factor. More...
static void	pitch_sharpening (AMRWBContext *ctx, float *fixed_vector)
	Apply pitch sharpening filters to the fixed codebook vector. More...
static float	voice_factor (float *p_vector, float p_gain, float *f_vector, float f_gain, CELPMContext *ctx)
	Calculate the voicing factor (-1.0 = unvoiced to 1.0 = voiced). More...
static float *	anti_sparseness (AMRWBContext *ctx, float *fixed_vector, float *buf)
	Reduce fixed vector sparseness by smoothing with one of three IR filters, also known as "adaptive phase dispersion". More...
static float	stability_factor (const float *isf, const float *isf_past)
	Calculate a stability factor {teta} based on distance between current and past isf. More...
static float	noise_enhancer (float fixed_gain, float *prev_tr_gain, float voice_fac, float stab_fac)
	Apply a non-linear fixed gain smoothing in order to reduce fluctuation in the energy of excitation. More...
static void	pitch_enhancer (float *fixed_vector, float voice_fac)
	Filter the fixed_vector to emphasize the higher frequencies. More...
static void	synthesis (AMRWBContext *ctx, float *lpc, float *excitation, float fixed_gain, const float *fixed_vector, float *samples)
	Conduct 16th order linear predictive coding synthesis from excitation. More...
static void	de_emphasis (float *out, float *in, float m, float mem[1])
	Apply to synthesis a de-emphasis filter of the form: H(z) = 1 / (1 - m * z^-1) More...
static void	upsample_5_4 (float *out, const float *in, int o_size, CELPMContext *ctx)
	Upsample a signal by 5/4 ratio (from 12.8kHz to 16kHz) using a FIR interpolation filter. More...
static float	find_hb_gain (AMRWBContext *ctx, const float *synth, uint16_t hb_idx, uint8_t vad)
	Calculate the high-band gain based on encoded index (23k85 mode) or on the low-band speech signal and the Voice Activity Detection flag. More...
static void	scaled_hb_excitation (AMRWBContext *ctx, float *hb_exc, const float *synth_exc, float hb_gain)
	Generate the high-band excitation with the same energy from the lower one and scaled by the given gain. More...
static float	auto_correlation (float *diff_isf, float mean, int lag)
	Calculate the auto-correlation for the ISF difference vector. More...
static void	extrapolate_isf (float isf[LP_ORDER_16k])
	Extrapolate a ISF vector to the 16kHz range (20th order LP) used at mode 6k60 LP filter for the high frequency band. More...
static void	lpc_weighting (float *out, const float *lpc, float gamma, int size)
	Spectral expand the LP coefficients using the equation: y[i] = x[i] * (gamma ** i) More...
static void	hb_synthesis (AMRWBContext *ctx, int subframe, float *samples, const float *exc, const float *isf, const float *isf_past)
	Conduct 20th order linear predictive coding synthesis for the high frequency band excitation at 16kHz. More...
static void	hb_fir_filter (float *out, const float fir_coef[HB_FIR_SIZE+1], float mem[HB_FIR_SIZE], const float *in)
	Apply a 15th order filter to high-band samples. More...
static void	update_sub_state (AMRWBContext *ctx)
	Update context state before the next subframe. More...
static int	amrwb_decode_frame (AVCodecContext *avctx, AVFrame *frame, int *got_frame_ptr, AVPacket *avpkt)

Variables
const FFCodec	ff_amrwb_decoder

Detailed Description

AMR wideband decoder

Definition in file amrwbdec.c.

Macro Definition Documentation

AMR_USE_16BIT_TABLES

#define AMR_USE_16BIT_TABLES

Definition at line 43 of file amrwbdec.c.

BIT_STR

#define BIT_STR	(		x,
			lsb,
			len
	)		av_mod_uintp2((x) >> (lsb), (len))

Get x bits in the index interval [lsb,lsb+len-1] inclusive.

Definition at line 372 of file amrwbdec.c.

BIT_POS

#define BIT_POS	(		x,
			p
	)		(((x) >> (p)) & 1)

Get the bit at specified position.

Definition at line 375 of file amrwbdec.c.

Function Documentation

amrwb_decode_init()

static av_cold int amrwb_decode_init ( AVCodecContext *  avctx )

static

Definition at line 104 of file amrwbdec.c.

decode_mime_header()

static int decode_mime_header ( AMRWBContext *  ctx, const uint8_t *  buf  )

static

Decode the frame header in the "MIME/storage" format.

This format is simpler and does not carry the auxiliary frame information.

Parameters

[in]	ctx	The Context
[in]	buf	Pointer to the input buffer

Returns

The decoded header length in bytes

Definition at line 154 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

decode_isf_indices_36b()

static void decode_isf_indices_36b ( uint16_t *  ind, float *  isf_q  )

static

Decode quantized ISF vectors using 36-bit indexes (6K60 mode only).

Parameters

[in]	ind	Array of 5 indexes
[out]	isf_q	Buffer for isf_q[LP_ORDER]

Definition at line 169 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

decode_isf_indices_46b()

static void decode_isf_indices_46b ( uint16_t *  ind, float *  isf_q  )

static

Decode quantized ISF vectors using 46-bit indexes (except 6K60 mode).

Parameters

[in]	ind	Array of 7 indexes
[out]	isf_q	Buffer for isf_q[LP_ORDER]

Definition at line 195 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

isf_add_mean_and_past()

static void isf_add_mean_and_past ( float *  isf_q, float *  isf_past  )

static

Apply mean and past ISF values using the prediction factor.

Updates past ISF vector.

Parameters

[in,out]	isf_q	Current quantized ISF
[in,out]	isf_past	Past quantized ISF

Definition at line 228 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

interpolate_isp()

static void interpolate_isp ( double  isp_q[4][LP_ORDER], const double *  isp4_past  )

static

Interpolate the fourth ISP vector from current and past frames to obtain an ISP vector for each subframe.

Parameters

[in,out]	isp_q	ISPs for each subframe
[in]	isp4_past	Past ISP for subframe 4

Definition at line 248 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

decode_pitch_lag_high()

static void decode_pitch_lag_high ( int *  lag_int, int *  lag_frac, int  pitch_index, uint8_t *  base_lag_int, int  subframe  )

static

Decode an adaptive codebook index into pitch lag (except 6k60, 8k85 modes).

Calculate integer lag and fractional lag always using 1/4 resolution. In 1st and 3rd subframes the index is relative to last subframe integer lag.

Parameters

[out]	lag_int	Decoded integer pitch lag
[out]	lag_frac	Decoded fractional pitch lag
[in]	pitch_index	Adaptive codebook pitch index
[in,out]	base_lag_int	Base integer lag used in relative subframes
[in]	subframe	Current subframe index (0 to 3)

Definition at line 270 of file amrwbdec.c.

Referenced by decode_pitch_vector().

decode_pitch_lag_low()

static void decode_pitch_lag_low ( int *  lag_int, int *  lag_frac, int  pitch_index, uint8_t *  base_lag_int, int  subframe, enum Mode  mode  )

static

Decode an adaptive codebook index into pitch lag for 8k85 and 6k60 modes.

The description is analogous to decode_pitch_lag_high, but in 6k60 the relative index is used for all subframes except the first.

Definition at line 303 of file amrwbdec.c.

Referenced by decode_pitch_vector().

decode_pitch_vector()

static void decode_pitch_vector ( AMRWBContext *  ctx, const AMRWBSubFrame *  amr_subframe, const int  subframe  )

static

Find the pitch vector by interpolating the past excitation at the pitch delay, which is obtained in this function.

Parameters

[in,out]	ctx	The context
[in]	amr_subframe	Current subframe data
[in]	subframe	Current subframe index (0 to 3)

Definition at line 332 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

decode_1p_track()

static void decode_1p_track ( int *  out, int  code, int  m, int  off  )

inlinestatic

The next six functions decode_[i]p_track decode exactly i pulses positions and amplitudes (-1 or 1) in a subframe track using an encoded pulse indexing (TS 26.190 section 5.8.2).

The results are given in out[], in which a negative number means amplitude -1 and vice versa (i.e., ampl(x) = x / abs(x) ).

Parameters

[out]	out	Output buffer (writes i elements)
[in]	code	Pulse index (no. of bits varies, see below)
[in]	m	(log2) Number of potential positions
[in]	off	Offset for decoded positions

code: m+1 bits

Definition at line 390 of file amrwbdec.c.

Referenced by decode_3p_track(), decode_4p_track(), decode_6p_track(), and decode_fixed_vector().

decode_2p_track()

static void decode_2p_track ( int *  out, int  code, int  m, int  off  )

inlinestatic

code: 2m+1 bits

Definition at line 397 of file amrwbdec.c.

Referenced by decode_3p_track(), decode_4p_track(), decode_5p_track(), decode_6p_track(), and decode_fixed_vector().

decode_3p_track()

static void decode_3p_track ( int *  out, int  code, int  m, int  off  )

static

code: 3m+1 bits

Definition at line 407 of file amrwbdec.c.

Referenced by decode_4p_track(), decode_5p_track(), decode_6p_track(), and decode_fixed_vector().

decode_4p_track()

static void decode_4p_track ( int *  out, int  code, int  m, int  off  )

static

code: 4m bits

Definition at line 416 of file amrwbdec.c.

Referenced by decode_6p_track(), and decode_fixed_vector().

decode_5p_track()

static void decode_5p_track ( int *  out, int  code, int  m, int  off  )

static

code: 5m bits

Definition at line 452 of file amrwbdec.c.

Referenced by decode_6p_track(), and decode_fixed_vector().

decode_6p_track()

static void decode_6p_track ( int *  out, int  code, int  m, int  off  )

static

code: 6m-2 bits

Definition at line 462 of file amrwbdec.c.

Referenced by decode_fixed_vector().

decode_fixed_vector()

static void decode_fixed_vector ( float *  fixed_vector, const uint16_t *  pulse_hi, const uint16_t *  pulse_lo, const enum Mode  mode  )

static

Decode the algebraic codebook index to pulse positions and signs, then construct the algebraic codebook vector.

Parameters

[out]	fixed_vector	Buffer for the fixed codebook excitation
[in]	pulse_hi	MSBs part of the pulse index array (higher modes only)
[in]	pulse_lo	LSBs part of the pulse index array
[in]	mode	Mode of the current frame

Definition at line 506 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

decode_gains()

static void decode_gains ( const uint8_t  vq_gain, const enum Mode  mode, float *  fixed_gain_factor, float *  pitch_gain  )

static

Decode pitch gain and fixed gain correction factor.

Parameters

[in]	vq_gain	Vector-quantized index for gains
[in]	mode	Mode of the current frame
[out]	fixed_gain_factor	Decoded fixed gain correction factor
[out]	pitch_gain	Decoded pitch gain

Definition at line 577 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

pitch_sharpening()

static void pitch_sharpening ( AMRWBContext *  ctx, float *  fixed_vector  )

static

Apply pitch sharpening filters to the fixed codebook vector.

Parameters

[in]	ctx	The context
[in,out]	fixed_vector	Fixed codebook excitation

Definition at line 595 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

voice_factor()

static float voice_factor ( float *  p_vector, float  p_gain, float *  f_vector, float  f_gain, CELPMContext *  ctx  )

static

Calculate the voicing factor (-1.0 = unvoiced to 1.0 = voiced).

Parameters

[in]	p_vector,f_vector	Pitch and fixed excitation vectors
[in]	p_gain,f_gain	Pitch and fixed gains
[in]	ctx	The context

Definition at line 617 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

anti_sparseness()

static float* anti_sparseness ( AMRWBContext *  ctx, float *  fixed_vector, float *  buf  )

static

Reduce fixed vector sparseness by smoothing with one of three IR filters, also known as "adaptive phase dispersion".

Parameters

[in]	ctx	The context
[in,out]	fixed_vector	Unfiltered fixed vector
[out]	buf	Space for modified vector if necessary

Returns

The potentially overwritten filtered fixed vector address

Definition at line 641 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

stability_factor()

static float stability_factor ( const float *  isf, const float *  isf_past  )

static

Calculate a stability factor {teta} based on distance between current and past isf.

A value of 1 shows maximum signal stability.

Definition at line 705 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

noise_enhancer()

static float noise_enhancer ( float  fixed_gain, float *  prev_tr_gain, float  voice_fac, float  stab_fac  )

static

Apply a non-linear fixed gain smoothing in order to reduce fluctuation in the energy of excitation.

Parameters

[in]	fixed_gain	Unsmoothed fixed gain
[in,out]	prev_tr_gain	Previous threshold gain (updated)
[in]	voice_fac	Frame voicing factor
[in]	stab_fac	Frame stability factor

Returns

The smoothed gain

Definition at line 729 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

pitch_enhancer()

static void pitch_enhancer ( float *  fixed_vector, float  voice_fac  )

static

Filter the fixed_vector to emphasize the higher frequencies.

Parameters

[in,out]	fixed_vector	Fixed codebook vector
[in]	voice_fac	Frame voicing factor

Definition at line 756 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

synthesis()

static void synthesis ( AMRWBContext *  ctx, float *  lpc, float *  excitation, float  fixed_gain, const float *  fixed_vector, float *  samples  )

static

Conduct 16th order linear predictive coding synthesis from excitation.

Parameters

[in]	ctx	Pointer to the AMRWBContext
[in]	lpc	Pointer to the LPC coefficients
[out]	excitation	Buffer for synthesis final excitation
[in]	fixed_gain	Fixed codebook gain for synthesis
[in]	fixed_vector	Algebraic codebook vector
[in,out]	samples	Pointer to the output samples and memory

Definition at line 784 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

de_emphasis()

static void de_emphasis ( float *  out, float *  in, float  m, float  mem[1]  )

static

Apply to synthesis a de-emphasis filter of the form: H(z) = 1 / (1 - m * z^-1)

Parameters

[out]	out	Output buffer
[in]	in	Input samples array with in[-1]
[in]	m	Filter coefficient
[in,out]	mem	State from last filtering

Definition at line 821 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

upsample_5_4()

static void upsample_5_4 ( float *  out, const float *  in, int  o_size, CELPMContext *  ctx  )

static

Upsample a signal by 5/4 ratio (from 12.8kHz to 16kHz) using a FIR interpolation filter.

Uses past data from before *in address.

Parameters

[out]	out	Buffer for interpolated signal
[in]	in	Current signal data (length 0.8*o_size)
[in]	o_size	Output signal length
[in]	ctx	The context

Definition at line 842 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

find_hb_gain()

static float find_hb_gain ( AMRWBContext *  ctx, const float *  synth, uint16_t  hb_idx, uint8_t  vad  )

static

Calculate the high-band gain based on encoded index (23k85 mode) or on the low-band speech signal and the Voice Activity Detection flag.

Parameters

[in]	ctx	The context
[in]	synth	LB speech synthesis at 12.8k
[in]	hb_idx	Gain index for mode 23k85 only
[in]	vad	VAD flag for the frame

Definition at line 874 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

scaled_hb_excitation()

static void scaled_hb_excitation ( AMRWBContext *  ctx, float *  hb_exc, const float *  synth_exc, float  hb_gain  )

static

Generate the high-band excitation with the same energy from the lower one and scaled by the given gain.

Parameters

[in]	ctx	The context
[out]	hb_exc	Buffer for the excitation
[in]	synth_exc	Low-band excitation used for synthesis
[in]	hb_gain	Wanted excitation gain

Definition at line 904 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

auto_correlation()

static float auto_correlation ( float *  diff_isf, float  mean, int  lag  )

static

Calculate the auto-correlation for the ISF difference vector.

Definition at line 923 of file amrwbdec.c.

Referenced by extrapolate_isf().

extrapolate_isf()

static void extrapolate_isf ( float  isf[LP_ORDER_16k] )

static

Extrapolate a ISF vector to the 16kHz range (20th order LP) used at mode 6k60 LP filter for the high frequency band.

Parameters

[out]

isf

Buffer for extrapolated isf; contains LP_ORDER values on input

Definition at line 942 of file amrwbdec.c.

Referenced by hb_synthesis().

lpc_weighting()

static void lpc_weighting ( float *  out, const float *  lpc, float  gamma, int  size  )

static

Spectral expand the LP coefficients using the equation: y[i] = x[i] * (gamma ** i)

Parameters

[out]	out	Output buffer (may use input array)
[in]	lpc	LP coefficients array
[in]	gamma	Weighting factor
[in]	size	LP array size

Definition at line 1007 of file amrwbdec.c.

Referenced by hb_synthesis().

hb_synthesis()

static void hb_synthesis ( AMRWBContext *  ctx, int  subframe, float *  samples, const float *  exc, const float *  isf, const float *  isf_past  )

static

Conduct 20th order linear predictive coding synthesis for the high frequency band excitation at 16kHz.

Parameters

[in]	ctx	The context
[in]	subframe	Current subframe index (0 to 3)
[in,out]	samples	Pointer to the output speech samples
[in]	exc	Generated white-noise scaled excitation
[in]	isf	Current frame isf vector
[in]	isf_past	Past frame final isf vector

Definition at line 1029 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

hb_fir_filter()

static void hb_fir_filter ( float *  out, const float  fir_coef[HB_FIR_SIZE+1], float  mem[HB_FIR_SIZE], const float *  in  )

static

Apply a 15th order filter to high-band samples.

The filter characteristic depends on the given coefficients.

Parameters

[out]	out	Buffer for filtered output
[in]	fir_coef	Filter coefficients
[in,out]	mem	State from last filtering (updated)
[in]	in	Input speech data (high-band)

Remarks

It is safe to pass the same array in in and out parameters

Definition at line 1070 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

update_sub_state()

static void update_sub_state ( AMRWBContext *  ctx )

static

Update context state before the next subframe.

Definition at line 1092 of file amrwbdec.c.

Referenced by amrwb_decode_frame().

amrwb_decode_frame()

static int amrwb_decode_frame ( AVCodecContext *  avctx, AVFrame *  frame, int *  got_frame_ptr, AVPacket *  avpkt  )

static

Definition at line 1108 of file amrwbdec.c.

Variable Documentation

ff_amrwb_decoder

const FFCodec ff_amrwb_decoder

Initial value:

= {
    .p.name         = "amrwb",
    CODEC_LONG_NAME("AMR-WB (Adaptive Multi-Rate WideBand)"),
    .p.type         = AVMEDIA_TYPE_AUDIO,
    .p.id           = AV_CODEC_ID_AMR_WB,
    .priv_data_size = sizeof(AMRWBChannelsContext),
    .init           = amrwb_decode_init,
    FF_CODEC_DECODE_CB(amrwb_decode_frame),
    .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_CHANNEL_CONF,
    .p.sample_fmts  = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
                                                     AV_SAMPLE_FMT_NONE },
}

Definition at line 1298 of file amrwbdec.c.