changeset 1305:9f9f55a8af92 mp3-gapless

Add gapless flag to MP3FileReader, and implement trimming the delay samples from the start (padding is not yet trimmed from end)

--- a/data/fileio/AudioFileReaderFactory.cpp	Tue Nov 29 08:58:50 2016 +0000
+++ b/data/fileio/AudioFileReaderFactory.cpp	Tue Nov 29 11:35:56 2016 +0000
@@ -165,7 +165,8 @@
 #ifdef HAVE_MAD
     if (!reader && MP3FileReader::supports(source)) {
         reader = new MP3FileReader
-            (source, decodeMode, cacheMode, targetRate, normalised, reporter);
+            (source, decodeMode, cacheMode, MP3FileReader::Gapless,
+             targetRate, normalised, reporter);
         CHECK(reader);
     }
 #endif
@@ -234,7 +235,8 @@
 #ifdef HAVE_MAD
     if (!reader) {
         reader = new MP3FileReader
-            (source, decodeMode, cacheMode, targetRate, normalised, reporter);
+            (source, decodeMode, cacheMode, MP3FileReader::Gapless,
+             targetRate, normalised, reporter);
         CHECK(reader);
     }
 #endif

--- a/data/fileio/CodedAudioFileReader.cpp	Tue Nov 29 08:58:50 2016 +0000
+++ b/data/fileio/CodedAudioFileReader.cpp	Tue Nov 29 11:35:56 2016 +0000
@@ -48,6 +48,8 @@
     m_normalised(normalised),
     m_max(0.f),
     m_gain(1.f),
+    m_trimFromStart(0),
+    m_trimFromEnd(0),
     m_clippedCount(0),
     m_firstNonzero(0),
     m_lastNonzero(0)
@@ -94,6 +96,13 @@
 }
 
 void
+CodedAudioFileReader::setSamplesToTrim(sv_frame_t fromStart, sv_frame_t fromEnd)
+{
+    m_trimFromStart = fromStart;
+    m_trimFromEnd = fromEnd;
+}
+
+void
 CodedAudioFileReader::startSerialised(QString id)
 {
     SVDEBUG << "CodedAudioFileReader(" << this << ")::startSerialised: id = " << id << endl;
@@ -223,6 +232,11 @@
     if (!m_initialised) return;
 
     for (sv_frame_t i = 0; i < nframes; ++i) {
+
+        if (m_trimFromStart > 0) {
+            --m_trimFromStart;
+            continue;
+        }
         
         for (int c = 0; c < m_channelCount; ++c) {
 
@@ -253,6 +267,11 @@
     if (!m_initialised) return;
 
     for (sv_frame_t i = 0; i < nframes; ++i) {
+
+        if (m_trimFromStart > 0) {
+            --m_trimFromStart;
+            continue;
+        }
         
         for (int c = 0; c < m_channelCount; ++c) {
 
@@ -283,6 +302,11 @@
     if (!m_initialised) return;
 
     for (float sample: samples) {
+
+        if (m_trimFromStart > 0) {
+            --m_trimFromStart;
+            continue;
+        }
         
         m_cacheWriteBuffer[m_cacheWriteBufferIndex++] = sample;
 
@@ -376,6 +400,7 @@
     float clip = 1.0;
     sv_frame_t count = sz * m_channelCount;
 
+    // statistics
     for (sv_frame_t j = 0; j < sz; ++j) {
         for (int c = 0; c < m_channelCount; ++c) {
             sv_frame_t i = j * m_channelCount + c;

--- a/data/fileio/CodedAudioFileReader.h	Tue Nov 29 08:58:50 2016 +0000
+++ b/data/fileio/CodedAudioFileReader.h	Tue Nov 29 11:35:56 2016 +0000
@@ -62,6 +62,9 @@
 
     void initialiseDecodeCache(); // samplerate, channels must have been set
 
+    // compensation for encoder delays:
+    void setSamplesToTrim(sv_frame_t fromStart, sv_frame_t fromEnd);
+    
     // may throw InsufficientDiscSpace:
     void addSamplesToDecodeCache(float **samples, sv_frame_t nframes);
     void addSamplesToDecodeCache(float *samplesInterleaved, sv_frame_t nframes);
@@ -104,6 +107,9 @@
     float m_max;
     float m_gain;
 
+    sv_frame_t m_trimFromStart;
+    sv_frame_t m_trimFromEnd;
+    
     sv_frame_t m_clippedCount;
     sv_frame_t m_firstNonzero;
     sv_frame_t m_lastNonzero;

--- a/data/fileio/MP3FileReader.cpp	Tue Nov 29 08:58:50 2016 +0000
+++ b/data/fileio/MP3FileReader.cpp	Tue Nov 29 11:35:56 2016 +0000
@@ -39,13 +39,19 @@
 #define open _open
 #endif
 
+using std::string;
+
+static sv_frame_t DEFAULT_DECODER_DELAY = 529;
+
 MP3FileReader::MP3FileReader(FileSource source, DecodeMode decodeMode, 
-                             CacheMode mode, sv_samplerate_t targetRate,
+                             CacheMode mode, GaplessMode gaplessMode,
+                             sv_samplerate_t targetRate,
                              bool normalised,
                              ProgressReporter *reporter) :
     CodedAudioFileReader(mode, targetRate, normalised),
     m_source(source),
     m_path(source.getLocalFilename()),
+    m_gaplessMode(gaplessMode),
     m_decodeErrorShown(false),
     m_decodeThread(0)
 {
@@ -65,6 +71,10 @@
     m_done = false;
     m_reporter = reporter;
 
+    if (m_gaplessMode == Gapless) {
+        CodedAudioFileReader::setSamplesToTrim(DEFAULT_DECODER_DELAY, 0);
+    }
+    
     struct stat stat;
     if (::stat(m_path.toLocal8Bit().data(), &stat) == -1 || stat.st_size == 0) {
 	m_error = QString("File %1 does not exist.").arg(m_path);
@@ -365,26 +375,92 @@
     return data->reader->filter(stream, frame);
 }
 
+static string toMagic(uint32_t fourcc)
+{
+    string magic("....");
+    for (int i = 0; i < 4; ++i) {
+        magic[3-i] = char((fourcc >> (8*i)) & 0xff);
+    }
+    return magic;
+}
+
 enum mad_flow
 MP3FileReader::filter(struct mad_stream const *stream,
                       struct mad_frame *)
 {
     if (m_mp3FrameCount > 0) {
+        // only handle info frame if it appears as first mp3 frame
         return MAD_FLOW_CONTINUE;
+    }
+
+    if (m_gaplessMode == Gappy) {
+        // Our non-gapless mode does not even filter out the Xing/LAME
+        // frame. That's because the main reason non-gapless mode
+        // exists is for backward compatibility with MP3FileReader
+        // behaviour before the gapless support was added, so we even
+        // need to keep the spurious 1152 samples resulting from
+        // feeding Xing/LAME frame to the decoder as otherwise we'd
+        // have different output from before.
+        SVDEBUG << "MP3FileReader: Not gapless mode, not checking Xing/LAME frame"
+                << endl;
+        return MAD_FLOW_CONTINUE;
+    }
+    
+    struct mad_bitptr ptr = stream->anc_ptr;
+    string magic = toMagic(mad_bit_read(&ptr, 32));
+
+    if (magic == "Xing" || magic == "Info") {
+
+        SVDEBUG << "MP3FileReader: Found Xing/LAME metadata frame (magic = \""
+                << magic << "\")" << endl;
+
+        // All we want at this point is the LAME encoder delay and
+        // padding values. We expect to see the Xing/Info magic (which
+        // we've already read), then 116 bytes of Xing data, then LAME
+        // magic, 5 byte version string, 12 bytes of LAME data that we
+        // aren't currently interested in, then the delays encoded as
+        // two 12-bit numbers into three bytes.
+        //
+        // (See gabriel.mp3-tech.org/mp3infotag.html)
+        
+        for (int skip = 0; skip < 116; ++skip) {
+            (void)mad_bit_read(&ptr, 8);
+        }
+
+        magic = toMagic(mad_bit_read(&ptr, 32));
+
+        if (magic == "LAME") {
+
+            SVDEBUG << "MP3FileReader: Found LAME-specific metadata" << endl;
+
+            for (int skip = 0; skip < 5 + 12; ++skip) {
+                (void)mad_bit_read(&ptr, 8);
+            }
+
+            uint32_t delay = mad_bit_read(&ptr, 12);
+            uint32_t padding = mad_bit_read(&ptr, 12);
+
+            sv_frame_t delayToDrop = DEFAULT_DECODER_DELAY + delay;
+            sv_frame_t paddingToDrop = padding - DEFAULT_DECODER_DELAY;
+            if (paddingToDrop < 0) paddingToDrop = 0;
+
+            SVDEBUG << "MP3FileReader: LAME encoder delay = " << delay
+                    << ", padding = " << padding << endl;
+
+            SVDEBUG << "MP3FileReader: Will be trimming " << delayToDrop
+                    << " samples at start and " << paddingToDrop
+                    << " at end" << endl;
+
+            CodedAudioFileReader::setSamplesToTrim(delayToDrop, paddingToDrop);
+            
+        } else {
+            SVDEBUG << "MP3FileReader: Xing frame has no LAME metadata" << endl;
+        }
+            
+        return MAD_FLOW_IGNORE;
+        
     } else {
-        struct mad_bitptr ptr = stream->anc_ptr;
-        unsigned long fourcc = mad_bit_read(&ptr, 32);
-        std::string magic("....");
-        for (int i = 0; i < 4; ++i) {
-            magic[3-i] = char((fourcc >> (8*i)) & 0xff);
-        }
-        if (magic == "Xing" || magic == "Info" || magic == "LAME") {
-            SVDEBUG << "MP3FileReader: Discarding metadata frame (magic = \""
-                    << magic << "\")" << endl;
-            return MAD_FLOW_IGNORE;
-        } else {
-            return MAD_FLOW_CONTINUE;
-        }
+        return MAD_FLOW_CONTINUE;
     }
 }
 

--- a/data/fileio/MP3FileReader.h	Tue Nov 29 08:58:50 2016 +0000
+++ b/data/fileio/MP3FileReader.h	Tue Nov 29 11:35:56 2016 +0000
@@ -32,9 +32,43 @@
     Q_OBJECT
 
 public:
+    /**
+     * How the MP3FileReader should handle leading and trailing gaps.
+     * See http://lame.sourceforge.net/tech-FAQ.txt for a technical
+     * explanation of the numbers here.
+     */
+    enum GaplessMode {
+        /**
+         * Trim unwanted samples from the start and end of the decoded
+         * audio. From the start, trim a number of samples equal to
+         * the decoder delay (a fixed 529 samples) plus any encoder
+         * delay that may be specified in Xing/LAME metadata. From the
+         * end, trim any padding specified in Xing/LAME metadata, less
+         * the fixed decoder delay. This usually results in "gapless"
+         * audio, i.e. with no spurious zero padding at either end.
+         */
+        Gapless,
+
+        /**
+         * Do not trim any samples. Also do not suppress any frames
+         * from being passed to the mp3 decoder, even Xing/LAME
+         * metadata frames. This will result in the audio being padded
+         * with zeros at either end: at the start, typically
+         * 529+576+1152 = 2257 samples for LAME-encoded mp3s; at the
+         * end an unknown number depending on the fill ratio of the
+         * final coded frame, but typically less than 1152-529 = 623.
+         *
+         * This mode produces the same output as produced by older
+         * versions of this code before the gapless option was added,
+         * and is present mostly for backward compatibility.
+         */
+        Gappy
+    };
+    
     MP3FileReader(FileSource source,
                   DecodeMode decodeMode,
                   CacheMode cacheMode,
+                  GaplessMode gaplessMode,
                   sv_samplerate_t targetRate = 0,
                   bool normalised = false,
                   ProgressReporter *reporter = 0);
@@ -68,6 +102,7 @@
     QString m_title;
     QString m_maker;
     TagMap m_tags;
+    GaplessMode m_gaplessMode;
     sv_frame_t m_fileSize;
     double m_bitrateNum;
     int m_bitrateDenom;