Merge pull request #9309 from kencooke/audio-hdr-volume

High Dynamic Range Audio injectors
This commit is contained in:
Brad Hefta-Gaub 2017-01-05 08:10:54 -08:00 committed by GitHub
commit 825fc1cfa1
7 changed files with 104 additions and 65 deletions

View file

@ -32,6 +32,7 @@
#include "AudioMixerClientData.h"
#include "AvatarAudioStream.h"
#include "InjectedAudioStream.h"
#include "AudioHelpers.h"
#include "AudioMixerSlave.h"
@ -406,63 +407,6 @@ void AudioMixerSlave::addStreamToMix(AudioMixerClientData& listenerNodeData, con
AudioConstants::NETWORK_FRAME_SAMPLES_PER_CHANNEL);
}
const int IEEE754_MANT_BITS = 23;
const int IEEE754_EXPN_BIAS = 127;
//
// for x > 0.0f, returns log2(x)
// for x <= 0.0f, returns large negative value
//
// abs |error| < 8e-3, smooth (exact for x=2^N) for NPOLY=3
// abs |error| < 2e-4, smooth (exact for x=2^N) for NPOLY=5
// rel |error| < 0.4 from precision loss very close to 1.0f
//
static inline float fastlog2(float x) {
union { float f; int32_t i; } mant, bits = { x };
// split into mantissa and exponent
mant.i = (bits.i & ((1 << IEEE754_MANT_BITS) - 1)) | (IEEE754_EXPN_BIAS << IEEE754_MANT_BITS);
int32_t expn = (bits.i >> IEEE754_MANT_BITS) - IEEE754_EXPN_BIAS;
mant.f -= 1.0f;
// polynomial for log2(1+x) over x=[0,1]
//x = (-0.346555386f * mant.f + 1.346555386f) * mant.f;
x = (((-0.0821307180f * mant.f + 0.321188984f) * mant.f - 0.677784014f) * mant.f + 1.43872575f) * mant.f;
return x + expn;
}
//
// for -126 <= x < 128, returns exp2(x)
//
// rel |error| < 3e-3, smooth (exact for x=N) for NPOLY=3
// rel |error| < 9e-6, smooth (exact for x=N) for NPOLY=5
//
static inline float fastexp2(float x) {
union { float f; int32_t i; } xi;
// bias such that x > 0
x += IEEE754_EXPN_BIAS;
//x = MAX(x, 1.0f);
//x = MIN(x, 254.9999f);
// split into integer and fraction
xi.i = (int32_t)x;
x -= xi.i;
// construct exp2(xi) as a float
xi.i <<= IEEE754_MANT_BITS;
// polynomial for exp2(x) over x=[0,1]
//x = (0.339766028f * x + 0.660233972f) * x + 1.0f;
x = (((0.0135557472f * x + 0.0520323690f) * x + 0.241379763f) * x + 0.693032121f) * x + 1.0f;
return x * xi.f;
}
float AudioMixerSlave::gainForSource(const AvatarAudioStream& listeningNodeStream, const PositionalAudioStream& streamToAdd,
const glm::vec3& relativePosition, bool isEcho) {
float gain = 1.0f;
@ -514,7 +458,7 @@ float AudioMixerSlave::gainForSource(const AvatarAudioStream& listeningNodeStrea
g = (g > 1.0f) ? 1.0f : g;
// calculate the distance coefficient using the distance to this node
float distanceCoefficient = fastexp2(fastlog2(g) * fastlog2(distanceBetween/ATTENUATION_BEGINS_AT_DISTANCE));
float distanceCoefficient = fastExp2f(fastLog2f(g) * fastLog2f(distanceBetween/ATTENUATION_BEGINS_AT_DISTANCE));
// multiply the current attenuation coefficient by the distance coefficient
gain *= distanceCoefficient;

View file

@ -25,6 +25,7 @@
#include "AudioLogging.h"
#include "SoundCache.h"
#include "AudioSRC.h"
#include "AudioHelpers.h"
int audioInjectorPtrMetaTypeId = qRegisterMetaType<AudioInjector*>();
@ -187,7 +188,7 @@ bool AudioInjector::injectLocally() {
return success;
}
const uchar MAX_INJECTOR_VOLUME = 0xFF;
const uchar MAX_INJECTOR_VOLUME = packFloatGainToByte(1.0f);
static const int64_t NEXT_FRAME_DELTA_ERROR_OR_FINISHED = -1;
static const int64_t NEXT_FRAME_DELTA_IMMEDIATELY = 0;
@ -333,7 +334,7 @@ int64_t AudioInjector::injectNextFrame() {
_currentPacket->writePrimitive(_options.position);
_currentPacket->writePrimitive(_options.orientation);
quint8 volume = MAX_INJECTOR_VOLUME * _options.volume;
quint8 volume = packFloatGainToByte(_options.volume);
_currentPacket->seek(volumeOptionOffset);
_currentPacket->writePrimitive(volume);

View file

@ -63,7 +63,7 @@ public:
AudioFOA& getLocalFOA() { return _localFOA; }
bool isLocalOnly() const { return _options.localOnly; }
float getVolume() const { return glm::clamp(_options.volume, 0.0f, 1.0f); }
float getVolume() const { return _options.volume; }
glm::vec3 getPosition() const { return _options.position; }
glm::quat getOrientation() const { return _options.orientation; }
bool isStereo() const { return _options.stereo; }

View file

@ -18,6 +18,7 @@
#include <UUID.h>
#include "InjectedAudioStream.h"
#include "AudioHelpers.h"
InjectedAudioStream::InjectedAudioStream(const QUuid& streamIdentifier, bool isStereo, int numStaticJitterFrames) :
PositionalAudioStream(PositionalAudioStream::Injector, isStereo, numStaticJitterFrames),
@ -25,8 +26,6 @@ InjectedAudioStream::InjectedAudioStream(const QUuid& streamIdentifier, bool isS
_radius(0.0f),
_attenuationRatio(0) {}
const uchar MAX_INJECTOR_VOLUME = 255;
int InjectedAudioStream::parseStreamProperties(PacketType type,
const QByteArray& packetAfterSeqNum,
int& numAudioSamples) {
@ -62,7 +61,7 @@ int InjectedAudioStream::parseStreamProperties(PacketType type,
quint8 attenuationByte = 0;
packetStream >> attenuationByte;
_attenuationRatio = attenuationByte / (float)MAX_INJECTOR_VOLUME;
_attenuationRatio = unpackFloatGainFromByte(attenuationByte);
packetStream >> _ignorePenumbra;

View file

@ -78,7 +78,7 @@ PacketVersion versionForPacketType(PacketType packetType) {
case PacketType::MicrophoneAudioNoEcho:
case PacketType::MicrophoneAudioWithEcho:
case PacketType::AudioStreamStats:
return static_cast<PacketVersion>(AudioVersion::SpaceBubbleChanges);
return static_cast<PacketVersion>(AudioVersion::HighDynamicRangeVolume);
default:
return 17;

View file

@ -245,6 +245,7 @@ enum class AudioVersion : PacketVersion {
TerminatingStreamStats,
SpaceBubbleChanges,
HasPersonalMute,
HighDynamicRangeVolume,
};
#endif // hifi_PacketHeaders_h

View file

@ -0,0 +1,94 @@
//
// AudioHelpers.h
// libraries/shared/src
//
// Created by Ken Cooke on 1/4/17.
// Copyright 2017 High Fidelity, Inc.
//
// Distributed under the Apache License, Version 2.0.
// See the accompanying file LICENSE or http://www.apache.org/licenses/LICENSE-2.0.html
//
#ifndef hifi_AudioHelpers_h
#define hifi_AudioHelpers_h
#include <stdint.h>
const int IEEE754_MANT_BITS = 23;
const int IEEE754_EXPN_BIAS = 127;
//
// for x > 0.0f, returns log2(x)
// for x <= 0.0f, returns large negative value
//
// abs |error| < 2e-4, smooth (exact for x=2^N)
// rel |error| < 0.4 from precision loss very close to 1.0f
//
static inline float fastLog2f(float x) {
union { float f; int32_t i; } mant, bits = { x };
// split into mantissa and exponent
mant.i = (bits.i & ((1 << IEEE754_MANT_BITS) - 1)) | (IEEE754_EXPN_BIAS << IEEE754_MANT_BITS);
int32_t expn = (bits.i >> IEEE754_MANT_BITS) - IEEE754_EXPN_BIAS;
mant.f -= 1.0f;
// polynomial for log2(1+x) over x=[0,1]
x = (((-0.0821307180f * mant.f + 0.321188984f) * mant.f - 0.677784014f) * mant.f + 1.43872575f) * mant.f;
return x + expn;
}
//
// for -127 <= x < 128, returns exp2(x)
// otherwise, returns undefined
//
// rel |error| < 9e-6, smooth (exact for x=N)
//
static inline float fastExp2f(float x) {
union { float f; int32_t i; } xi;
// bias such that x > 0
x += IEEE754_EXPN_BIAS;
// split into integer and fraction
xi.i = (int32_t)x;
x -= xi.i;
// construct exp2(xi) as a float
xi.i <<= IEEE754_MANT_BITS;
// polynomial for exp2(x) over x=[0,1]
x = (((0.0135557472f * x + 0.0520323690f) * x + 0.241379763f) * x + 0.693032121f) * x + 1.0f;
return x * xi.f;
}
//
// Quantize a non-negative gain value to the nearest 0.5dB, and pack to a byte.
//
// Values above +30dB are clamped to +30dB
// Values below -97dB are clamped to -inf
// Value of 1.0 (+0dB) is reconstructed exactly
//
const float GAIN_CONVERSION_RATIO = 2.0f * 6.02059991f; // scale log2 to 0.5dB
const float GAIN_CONVERSION_OFFSET = 255 - 60.0f; // translate +30dB to max
static inline uint8_t packFloatGainToByte(float gain) {
float f = fastLog2f(gain) * GAIN_CONVERSION_RATIO + GAIN_CONVERSION_OFFSET;
int32_t i = (int32_t)(f + 0.5f); // quantize
uint8_t byte = (i < 0) ? 0 : ((i > 255) ? 255 : i); // clamp
return byte;
}
static inline float unpackFloatGainFromByte(uint8_t byte) {
float gain = (byte == 0) ? 0.0f : fastExp2f((byte - GAIN_CONVERSION_OFFSET) * (1.0f/GAIN_CONVERSION_RATIO));
return gain;
}
#endif // hifi_AudioHelpers_h