Mercurial > hg > svcore

changeset 759:a43acbe3988f

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
More refinement in audiofile read tests and implementation
author Chris Cannam
date Fri, 08 Mar 2013 21:35:46 +0000
parents babed5be1ae7
children b6bb0ecb7958
files data/fileio/CodedAudioFileReader.cpp data/fileio/CoreAudioFileReader.cpp data/fileio/test/AudioFileReaderTest.h data/fileio/test/testfiles/32000-1.aac data/fileio/test/testfiles/32000-1.m4a data/fileio/test/testfiles/44100-2.aac data/fileio/test/testfiles/44100-2.m4a
diffstat 7 files changed, 77 insertions(+), 32 deletions(-) [+]
line wrap: on
line diff
--- a/data/fileio/CodedAudioFileReader.cpp	Fri Mar 08 20:05:49 2013 +0000
+++ b/data/fileio/CodedAudioFileReader.cpp	Fri Mar 08 21:35:46 2013 +0000
@@ -270,11 +270,11 @@
         return;
     }
 
-    if (m_cacheWriteBufferIndex > 0) {
+//    if (m_cacheWriteBufferIndex > 0) {
         pushBuffer(m_cacheWriteBuffer,
                    m_cacheWriteBufferIndex / m_channelCount,
                    true);
-    }        
+//    }        
 
     delete[] m_cacheWriteBuffer;
     m_cacheWriteBuffer = 0;
@@ -349,14 +349,19 @@
 CodedAudioFileReader::pushBufferResampling(float *buffer, size_t sz,
                                            float ratio, bool final)
 {
-    size_t out = m_resampler->resampleInterleaved
-        (buffer,
-         m_resampleBuffer,
-         sz,
-         ratio,
-         false);
+    SVDEBUG << "pushBufferResampling: ratio = " << ratio << ", sz = " << sz << ", final = " << final << endl;
 
-    pushBufferNonResampling(m_resampleBuffer, out);
+    if (sz > 0) {
+
+        size_t out = m_resampler->resampleInterleaved
+            (buffer,
+             m_resampleBuffer,
+             sz,
+             ratio,
+             false);
+
+        pushBufferNonResampling(m_resampleBuffer, out);
+    }
 
     if (final) {
 
@@ -367,18 +372,22 @@
 
         size_t padSamples = padFrames * m_channelCount;
 
-        std::cerr << "frameCount = " << m_frameCount << ", equivFileFrames = " << m_frameCount / ratio << ", m_fileFrameCount = " << m_fileFrameCount << ", padFrames= " << padFrames << ", padSamples = " << padSamples << std::endl;
+        SVDEBUG << "frameCount = " << m_frameCount << ", equivFileFrames = " << m_frameCount / ratio << ", m_fileFrameCount = " << m_fileFrameCount << ", padFrames= " << padFrames << ", padSamples = " << padSamples << endl;
 
         float *padding = new float[padSamples];
         for (int i = 0; i < padSamples; ++i) padding[i] = 0.f;
 
-        out = m_resampler->resampleInterleaved
+        size_t out = m_resampler->resampleInterleaved
             (padding,
              m_resampleBuffer,
              padFrames,
              ratio,
              true);
 
+        if (m_frameCount + out > int(m_fileFrameCount * ratio)) {
+            out = int(m_fileFrameCount * ratio) - m_frameCount;
+        }
+
         pushBufferNonResampling(m_resampleBuffer, out);
         delete[] padding;
     }
--- a/data/fileio/CoreAudioFileReader.cpp	Fri Mar 08 20:05:49 2013 +0000
+++ b/data/fileio/CoreAudioFileReader.cpp	Fri Mar 08 21:35:46 2013 +0000
@@ -74,7 +74,7 @@
 
     Profiler profiler("CoreAudioFileReader::CoreAudioFileReader", true);
 
-    std::cerr << "CoreAudioFileReader: path is \"" << m_path.toStdString() << "\"" << std::endl;
+    SVDEBUG << "CoreAudioFileReader: path is \"" << m_path.toStdString() << "\"" << endl;
 
     QByteArray ba = m_path.toLocal8Bit();
 
@@ -120,9 +120,7 @@
     m_channelCount = m_d->asbd.mChannelsPerFrame;
     m_fileRate = m_d->asbd.mSampleRate;
 
-    std::cerr << "CoreAudioReadStream: " << m_channelCount << " channels, " << m_sampleRate << " Hz" << std::endl;
-
-    m_d->asbd.mSampleRate = getSampleRate();
+    std::cerr << "CoreAudioReadStream: " << m_channelCount << " channels, " << m_fileRate << " Hz" << std::endl;
 
     m_d->asbd.mFormatID = kAudioFormatLinearPCM;
     m_d->asbd.mFormatFlags =
--- a/data/fileio/test/AudioFileReaderTest.h	Fri Mar 08 20:05:49 2013 +0000
+++ b/data/fileio/test/AudioFileReaderTest.h	Fri Mar 08 21:35:46 2013 +0000
@@ -89,22 +89,24 @@
 	AudioTestData tdata(readRate, channels);
 	
 	float *reference = tdata.getInterleavedData();
-	int refsize = tdata.getFrameCount() * channels;
+        int refFrames = tdata.getFrameCount();
+	int refsize = refFrames * channels;
 	
 	vector<float> test;
 	
 	// The reader should give us exactly the expected number of
-	// frames, except for mp3 files -- so we ask for one more,
-	// just to check we don't get it!
-	reader->getInterleavedFrames
-	    (0, tdata.getFrameCount() + 1, test);
+	// frames, except for mp3/aac files. We ask for quite a lot
+	// more, though, so we can (a) check that we only get the
+	// expected number back (if this is not mp3/aac) or (b) take
+	// into account silence at beginning and end (if it is).
+	reader->getInterleavedFrames(0, refFrames + 5000, test);
 	int read = test.size() / channels;
 
-        if (extension == "mp3") {
-            // mp3s round up
-            QVERIFY(read >= tdata.getFrameCount());
+        if (extension == "mp3" || extension == "aac" || extension == "m4a") {
+            // mp3s and aacs can have silence at start and end
+            QVERIFY(read >= refFrames);
         } else {
-            QCOMPARE(read, tdata.getFrameCount());
+            QCOMPARE(read, refFrames);
         }
 
         // Our limits are pretty relaxed -- we're not testing decoder
@@ -112,26 +114,58 @@
         // plainly wrong (e.g. at wrong samplerate or with an offset)
 
 	float limit = 0.01;
+        float edgeLimit = limit * 10; // in first or final edgeSize frames
+        int edgeSize = 100; 
+
         if (nominalDepth < 16) {
             limit = 0.02;
         }
-        if (extension == "ogg" || extension == "mp3" || extension == "aac") {
-            limit = 0.04;
+        if (extension == "ogg" || extension == "mp3" ||
+            extension == "aac" || extension == "m4a") {
+            limit = 0.2;
+            edgeLimit = limit * 3;
         }
 
-        int edgeSize = 100; 
-        float edgeLimit = limit * 10; // in first or final edgeSize frames
+        // And we ignore completely the last few frames when upsampling
+        int discard = 1 + readRate / nominalRate;
+
+        int offset = 0;
+
+        if (extension == "aac" || extension == "m4a") {
+            // our m4a file appears to have a fixed offset of 1024 (at
+            // file sample rate)
+            offset = (1024 / float(nominalRate)) * readRate;
+        }
+
+        if (extension == "mp3") {
+            // while mp3s appear to vary
+            for (int i = 0; i < read; ++i) {
+                bool any = false;
+                float thresh = 0.01;
+                for (int c = 0; c < channels; ++c) {
+                    if (fabsf(test[i * channels + c]) > thresh) {
+                        any = true;
+                        break;
+                    }
+                }
+                if (any) {
+                    offset = i;
+                    break;
+                }
+            }
+//            std::cerr << "offset = " << offset << std::endl;
+        }
 
 	for (int c = 0; c < channels; ++c) {
 	    float maxdiff = 0.f;
 	    int maxAt = 0;
 	    float totdiff = 0.f;
-	    for (int i = 0; i < read; ++i) {
-		float diff = fabsf(test[i * channels + c] -
+	    for (int i = 0; i < read - offset - discard && i < refFrames; ++i) {
+		float diff = fabsf(test[(i + offset) * channels + c] -
 				   reference[i * channels + c]);
 		totdiff += diff;
                 // in edge areas, record this only if it exceeds edgeLimit
-                if (i < edgeSize || i + edgeSize >= read) {
+                if (i < edgeSize || i + edgeSize >= read - offset) {
                     if (diff > edgeLimit) {
                         maxdiff = diff;
                         maxAt = i;
@@ -146,8 +180,12 @@
 	    float meandiff = totdiff / read;
 //	    cerr << "meandiff on channel " << c << ": " << meandiff << endl;
 //	    cerr << "maxdiff on channel " << c << ": " << maxdiff << " at " << maxAt << endl;
+            if (meandiff >= limit) {
+		cerr << "ERROR: for audiofile " << audiofile << ": mean diff = " << meandiff << " for channel " << c << endl;
+                QVERIFY(meandiff < limit);
+            }
 	    if (maxdiff >= limit) {
-		cerr << "ERROR: for audiofile " << audiofile << ": maxdiff = " << maxdiff << " at frame " << maxAt << " of " << read << " on channel " << c << " (mean diff = " << meandiff << ")" << endl;
+		cerr << "ERROR: for audiofile " << audiofile << ": max diff = " << maxdiff << " at frame " << maxAt << " of " << read << " on channel " << c << " (mean diff = " << meandiff << ")" << endl;
 		QVERIFY(maxdiff < limit);
 	    }
 	}
Binary file data/fileio/test/testfiles/32000-1.aac has changed
Binary file data/fileio/test/testfiles/32000-1.m4a has changed
Binary file data/fileio/test/testfiles/44100-2.aac has changed
Binary file data/fileio/test/testfiles/44100-2.m4a has changed