@ -1,6 +1,5 @@
package com.otaliastudios.cameraview.video.encoding ;
import android.media.AudioFormat ;
import android.media.AudioRecord ;
import android.media.MediaCodec ;
import android.media.MediaCodecInfo ;
@ -15,8 +14,10 @@ import androidx.annotation.RequiresApi;
import java.io.IOException ;
import java.nio.ByteBuffer ;
import java.nio.ByteOrder ;
import java.util.HashMap ;
import java.util.Map ;
import java.util.Random ;
import java.util.concurrent.LinkedBlockingQueue ;
/ * *
@ -30,23 +31,24 @@ public class AudioMediaEncoder extends MediaEncoder {
private static final boolean PERFORMANCE_DEBUG = false ;
private static final boolean PERFORMANCE_FILL_GAPS = true ;
private static final int PERFORMANCE_MAX_GAPS = 8 ;
private boolean mRequestStop = false ;
private AudioEncodingThread mEncoder ;
private AudioRecordingThread mRecorder ;
private ByteBufferPool mByteBufferPool ;
private ByteBuffer mZeroBuffer ;
private final AudioTimestamp mTimestamp ;
private AudioConfig mConfig ;
private InputBufferPool mInputBufferPool = new InputBufferPool ( ) ;
private final LinkedBlockingQueue < InputBuffer > mInputBufferQueue = new LinkedBlockingQueue < > ( ) ;
private AudioNoise mAudioNoise ;
// Just to debug performance.
private int mSendCount = 0 ;
private int mExecuteCount = 0 ;
private long mAvgSend Delay = 0 ;
private long mAvgExecute Delay = 0 ;
private Map < Long , Long > mSendStartMap = new HashMap < > ( ) ;
private int mDebug SendCount = 0 ;
private int mDebug ExecuteCount = 0 ;
private long mDebugSendAvg Delay = 0 ;
private long mDebugExecuteAvg Delay = 0 ;
private Map < Long , Long > mDebug SendStartMap = new HashMap < > ( ) ;
public AudioMediaEncoder ( @NonNull AudioConfig config ) {
super ( "AudioEncoder" ) ;
@ -76,7 +78,7 @@ public class AudioMediaEncoder extends MediaEncoder {
mMediaCodec . configure ( audioFormat , null , null , MediaCodec . CONFIGURE_FLAG_ENCODE ) ;
mMediaCodec . start ( ) ;
mByteBufferPool = new ByteBufferPool ( mConfig . frameSize ( ) , mConfig . bufferPoolMaxSize ( ) ) ;
mZeroBuffer = ByteBuffer . allocateDirect ( mConfig . frameSize ( ) ) ;
mAudioNoise = new AudioNoise ( mConfig ) ;
}
@EncoderThread
@ -130,11 +132,13 @@ public class AudioMediaEncoder extends MediaEncoder {
private AudioRecord mAudioRecord ;
private ByteBuffer mCurrentBuffer ;
private int mReadBytes ;
private int mCurrentReadBytes ;
private long mLastTimeUs ;
private long mFirstTimeUs = Long . MIN_VALUE ;
private AudioRecordingThread ( ) {
setPriority ( Thread . MAX_PRIORITY ) ;
final int minBufferSize = AudioRecord . getMinBufferSize (
mConfig . samplingFrequency ,
mConfig . audioFormatChannels ( ) ,
@ -152,14 +156,22 @@ public class AudioMediaEncoder extends MediaEncoder {
mConfig . audioFormatChannels ( ) ,
mConfig . encoding ,
bufferSize ) ;
setPriority ( Thread . MAX_PRIORITY ) ;
}
@Override
public void run ( ) {
mAudioRecord . startRecording ( ) ;
while ( ! mRequestStop ) {
if ( ! hasReachedMaxLength ( ) ) {
read ( false ) ;
} else {
// We have reached the max length, so stop reading.
// However, do not get out of the loop - the controller
// will call stop() on us soon. It's not our responsibility
// to stop ourselves.
//noinspection UnnecessaryContinue
continue ;
}
}
LOG . w ( "Stop was requested. We're out of the loop. Will post an endOfStream." ) ;
// Last input with 0 length. This will signal the endOfStream.
@ -192,25 +204,25 @@ public class AudioMediaEncoder extends MediaEncoder {
// with left and right bytes. https://stackoverflow.com/q/20594750/4288782
if ( PERFORMANCE_DEBUG ) {
long before = System . nanoTime ( ) ;
mReadBytes = mAudioRecord . read ( mCurrentBuffer , mConfig . frameSize ( ) ) ;
mCurrent ReadBytes = mAudioRecord . read ( mCurrentBuffer , mConfig . frameSize ( ) ) ;
long after = System . nanoTime ( ) ;
float delayMillis = ( after - before ) / 1000000F ;
float durationMillis = AudioTimestamp . bytesToMillis ( mReadBytes , mConfig . byteRate ( ) ) ;
float durationMillis = AudioTimestamp . bytesToMillis ( mCurrent ReadBytes , mConfig . byteRate ( ) ) ;
LOG . v ( "read thread - reading took:" , delayMillis ,
"should be:" , durationMillis ,
"delay:" , delayMillis - durationMillis ) ;
} else {
mReadBytes = mAudioRecord . read ( mCurrentBuffer , mConfig . frameSize ( ) ) ;
mCurrent ReadBytes = mAudioRecord . read ( mCurrentBuffer , mConfig . frameSize ( ) ) ;
}
LOG . i ( "read thread - eos:" , endOfStream , "- Read new audio frame. Bytes:" , mReadBytes ) ;
if ( mReadBytes > 0 ) { // Good read: increase PTS.
increaseTime ( mReadBytes , endOfStream ) ;
LOG . i ( "read thread - eos:" , endOfStream , "- Read new audio frame. Bytes:" , mCurrent ReadBytes ) ;
if ( mCurrent ReadBytes > 0 ) { // Good read: increase PTS.
increaseTime ( mCurrent ReadBytes , endOfStream ) ;
LOG . i ( "read thread - eos:" , endOfStream , "- mLastTimeUs:" , mLastTimeUs ) ;
mCurrentBuffer . limit ( mReadBytes ) ;
mCurrentBuffer . limit ( mCurrent ReadBytes ) ;
enqueue ( mCurrentBuffer , mLastTimeUs , endOfStream ) ;
} else if ( mReadBytes = = AudioRecord . ERROR_INVALID_OPERATION ) {
} else if ( mCurrent ReadBytes = = AudioRecord . ERROR_INVALID_OPERATION ) {
LOG . e ( "read thread - eos:" , endOfStream , "- Got AudioRecord.ERROR_INVALID_OPERATION" ) ;
} else if ( mReadBytes = = AudioRecord . ERROR_BAD_VALUE ) {
} else if ( mCurrent ReadBytes = = AudioRecord . ERROR_BAD_VALUE ) {
LOG . e ( "read thread - eos:" , endOfStream , "- Got AudioRecord.ERROR_BAD_VALUE" ) ;
}
}
@ -235,43 +247,21 @@ public class AudioMediaEncoder extends MediaEncoder {
}
// See if we reached the max length value.
if ( ! hasReachedMaxLength ( ) ) {
boolean didReachMaxLength = ( mLastTimeUs - mFirstTimeUs ) > getMaxLengthMillis ( ) * 1000L ;
if ( didReachMaxLength & & ! endOfStream ) {
LOG . w ( "read thread - this frame reached the maxLength! deltaUs:" , mLastTimeUs - mFirstTimeUs ) ;
notifyMaxLengthReached ( ) ;
}
// Add zeroes if we have huge gaps. Even if timestamps are correct, if we have gaps between
// them, the encoder might shrink all timestamps to have a continuous audio. This results
// in a video that is fast-forwarded.
// Adding zeroes does not solve the gaps issue - audio will still be distorted. But at
// least we get a video that has the correct playback speed.
if ( PERFORMANCE_FILL_GAPS ) {
int gaps = mTimestamp . getGapCount ( mConfig . frameSize ( ) ) ;
if ( gaps > 0 ) {
long gapStart = mTimestamp . getGapStartUs ( mLastTimeUs ) ;
long frameUs = AudioTimestamp . bytesToUs ( mConfig . frameSize ( ) , mConfig . byteRate ( ) ) ;
LOG . w ( "read thread - GAPS: trying to add" , gaps , "zeroed buffers" ) ;
for ( int i = 0 ; i < gaps ; i + + ) {
ByteBuffer zeroBuffer = mByteBufferPool . get ( ) ;
if ( zeroBuffer = = null ) {
LOG . e ( "read thread - GAPS: aborting because we have no free buffer." ) ;
break ;
}
;
zeroBuffer . position ( 0 ) ;
zeroBuffer . put ( mZeroBuffer ) ;
zeroBuffer . clear ( ) ;
enqueue ( zeroBuffer , gapStart , false ) ;
gapStart + = frameUs ;
}
}
}
// Maybe add noise.
maybeAddNoise ( ) ;
}
private void enqueue ( @NonNull ByteBuffer byteBuffer , long timestamp , boolean isEndOfStream ) {
if ( PERFORMANCE_DEBUG ) {
mSendStartMap . put ( timestamp , System . nanoTime ( ) / 1000000 ) ;
mDebugSendStartMap . put ( timestamp , System . nanoTime ( ) / 1000000 ) ;
}
int readBytes = byteBuffer . remaining ( ) ;
InputBuffer inputBuffer = mInputBufferPool . get ( ) ;
@ -283,6 +273,45 @@ public class AudioMediaEncoder extends MediaEncoder {
mInputBufferQueue . add ( inputBuffer ) ;
}
/ * *
* If our { @link AudioTimestamp } detected huge gap , and the performance flag is enabled ,
* we can add noise to fill them .
*
* Even if we always pass the correct timestamps , if there are big gaps between the frames ,
* the encoder implementation might shrink all timestamps to have a continuous audio .
* This results in a video that is fast - forwarded .
*
* Adding noise does not solve the gaps issue , we ' ll still have distorted audio , but
* at least we get a video that has the correct playback speed .
*
* NOTE : this MUST be fast !
* If this operation is slow , we make the { @link AudioRecordingThread } busy , so we ' ll
* read the next frame with a delay , so we ' ll have even more gaps at the next call
* and spend even more time here . The result might be recording no audio at all - just
* random noise .
* This is the reason why we have a { @link # PERFORMANCE_MAX_GAPS } number .
* /
private void maybeAddNoise ( ) {
if ( ! PERFORMANCE_FILL_GAPS ) return ;
int gaps = mTimestamp . getGapCount ( mConfig . frameSize ( ) ) ;
if ( gaps < = 0 ) return ;
long gapStart = mTimestamp . getGapStartUs ( mLastTimeUs ) ;
long frameUs = AudioTimestamp . bytesToUs ( mConfig . frameSize ( ) , mConfig . byteRate ( ) ) ;
LOG . w ( "read thread - GAPS: trying to add" , gaps , "noise buffers. PERFORMANCE_MAX_GAPS:" , PERFORMANCE_MAX_GAPS ) ;
for ( int i = 0 ; i < Math . min ( gaps , PERFORMANCE_MAX_GAPS ) ; i + + ) {
ByteBuffer noiseBuffer = mByteBufferPool . get ( ) ;
if ( noiseBuffer = = null ) {
LOG . e ( "read thread - GAPS: aborting because we have no free buffer." ) ;
break ;
}
noiseBuffer . clear ( ) ;
mAudioNoise . fill ( noiseBuffer ) ;
noiseBuffer . rewind ( ) ;
enqueue ( noiseBuffer , gapStart , false ) ;
gapStart + = frameUs ;
}
}
}
/ * *
@ -311,10 +340,11 @@ public class AudioMediaEncoder extends MediaEncoder {
// Performance logging
if ( PERFORMANCE_DEBUG ) {
long sendEnd = System . nanoTime ( ) / 1000000 ;
Long sendStart = mSendStartMap . remove ( inputBuffer . timestamp ) ;
Long sendStart = mDebugSendStartMap . remove ( inputBuffer . timestamp ) ;
//noinspection StatementWithEmptyBody
if ( sendStart ! = null ) {
mAvgSend Delay = ( ( mAvgSend Delay * mSendCount ) + ( sendEnd - sendStart ) ) / ( + + mSendCount ) ;
LOG . v ( "send delay millis:" , sendEnd - sendStart , "average:" , mAvgSend Delay ) ;
mDebugSendAvg Delay = ( ( mDebugSendAvg Delay * mDebug SendCount ) + ( sendEnd - sendStart ) ) / ( + + mDebug SendCount ) ;
LOG . v ( "send delay millis:" , sendEnd - sendStart , "average:" , mDebugSendAvg Delay ) ;
} else {
// This input buffer was already processed (but tryAcquire failed for now).
}
@ -338,8 +368,8 @@ public class AudioMediaEncoder extends MediaEncoder {
if ( PERFORMANCE_DEBUG ) {
// After latest changes, the count here is not so different between MONO and STEREO.
// We get about 400 frames in both cases (430 for MONO, but doesn't seem like a big issue).
LOG . e ( "EXECUTE DELAY MILLIS:" , mAvgExecute Delay , "COUNT:" , mExecuteCount ) ;
LOG . e ( "SEND DELAY MILLIS:" , mAvgSend Delay , "COUNT:" , mSendCount ) ;
LOG . e ( "EXECUTE DELAY MILLIS:" , mDebugExecuteAvg Delay , "COUNT:" , mDebug ExecuteCount ) ;
LOG . e ( "SEND DELAY MILLIS:" , mDebugSendAvg Delay , "COUNT:" , mDebug SendCount ) ;
}
}
@ -357,12 +387,12 @@ public class AudioMediaEncoder extends MediaEncoder {
// NOTE: can consider calling this drainOutput on yet another thread, which would let us
// use an even smaller BUFFER_POOL_MAX_SIZE without losing audio frames. But this way
// we can accumulate delay on this new thread without noticing (no pool getting empty).
drainOutput ( buff er . isEndOfStream ) ;
drainOutput ( eos ) ;
if ( PERFORMANCE_DEBUG ) {
long executeEnd = System . nanoTime ( ) / 1000000 ;
mAvgExecute Delay = ( ( mAvgExecute Delay * mExecuteCount ) + ( executeEnd - executeStart ) ) / ( + + mExecuteCount ) ;
LOG . v ( "execute delay millis:" , executeEnd - executeStart , "average:" , mAvgExecute Delay ) ;
mDebugExecuteAvg Delay = ( ( mDebugExecuteAvg Delay * mDebug ExecuteCount ) + ( executeEnd - executeStart ) ) / ( + + mDebug ExecuteCount ) ;
LOG . v ( "execute delay millis:" , executeEnd - executeStart , "average:" , mDebugExecuteAvg Delay ) ;
}
}
}
Mattia Iavarone
5 years ago
committed by
GitHub