Merge pull request wang-bin#970 from 0xFelix/splitAudioFrames

Split audio frames

src/AudioFormat.cpp

@@ -31,8 +31,6 @@
 
 namespace QtAV {
 
-const qint64 kHz = 1000000LL;
-
 typedef struct {
     AVSampleFormat avfmt;
     AudioFormat::SampleFormat fmt;

src/AudioFrame.cpp

@@ -132,27 +132,72 @@ int AudioFrame::channelCount() const
 }
 
 AudioFrame AudioFrame::clone() const
+{
+    return mid(0, -1);
+}
+
+AudioFrame AudioFrame::mid(int pos, int len) const
 {
     Q_D(const AudioFrame);
+
     if (d->format.sampleFormatFFmpeg() == AV_SAMPLE_FMT_NONE
-            || d->format.channels() <= 0)
+            || d->format.channels() <= 0) {
         return AudioFrame();
-    if (d->samples_per_ch <= 0 || bytesPerLine(0) <= 0)
+    }
+
+    if (d->samples_per_ch <= 0 || bytesPerLine(0) <= 0 || len == 0) {
         return AudioFrame(format());
-    QByteArray buf(bytesPerLine()*planeCount(), 0);
+    }
+
+    int bufSize = bytesPerLine();
+    int posBytes = 0;
+    if (pos > 0) {
+        posBytes = pos * d->format.bytesPerSample();
+        bufSize -= posBytes;
+    } else {
+        pos = 0;
+    }
+
+    int lenBytes = len * d->format.bytesPerSample();
+    if (len > 0 && lenBytes < bufSize) {
+        bufSize = lenBytes;
+    } else {
+        lenBytes = bufSize;
+    }
+
+    QByteArray buf(bufSize * planeCount(), 0);
     char *dst = buf.data(); //must before buf is shared, otherwise data will be detached.
+
     for (int i = 0; i < planeCount(); ++i) {
-        const int plane_size = bytesPerLine(i);
-        memcpy(dst, constBits(i), plane_size);
-        dst += plane_size;
+        memcpy(dst, constBits(i) + posBytes, lenBytes);
+        dst += lenBytes;
     }
+
     AudioFrame f(d->format, buf);
-    f.setSamplesPerChannel(samplesPerChannel());
-    f.setTimestamp(timestamp());
+    f.setSamplesPerChannel(bufSize / d->format.bytesPerSample());
+    f.setTimestamp(d->timestamp + (qreal) d->format.durationForBytes(posBytes) / AudioFormat::kHz);
     // meta data?
     return f;
 }
 
+void AudioFrame::prepend(AudioFrame &other)
+{
+    Q_D(AudioFrame);
+
+    if (d->format != other.format()) {
+        qWarning() << "To prepend a frame it must have the same audio format";
+        return;
+    }
+
+    d->data.prepend(other.data());
+    d->samples_per_ch += other.samplesPerChannel();
+    d->timestamp = other.timestamp();
+
+    for (int i = 0; i < planeCount(); i++) {
+        d->line_sizes[i] += other.bytesPerLine(i);
+    }
+}
+
 AudioFormat AudioFrame::format() const
 {
     return d_func()->format;

src/QtAV/AudioEncoder.h

@@ -64,6 +64,8 @@ class Q_AV_EXPORT AudioEncoder : public AVEncoder
      */
     const AudioFormat& audioFormat() const;
     void setAudioFormat(const AudioFormat& format);
+
+    int frameSize() const;
 Q_SIGNALS:
     void audioFormatChanged();
 public:

src/QtAV/AudioFormat.h

@@ -68,6 +68,9 @@ class Q_AV_EXPORT AudioFormat
         ChannelLayout_Stereo,
         ChannelLayout_Unsupported //ok. now it's not complete
     };
+
+    static const qint64 kHz = 1000000LL;
+
     //typedef qint64 ChannelLayout; //currently use latest FFmpeg's
     // TODO: constexpr
     friend int RawSampleSize(SampleFormat fmt) { return fmt & ((1<<(kSize+1)) - 1); }

src/QtAV/AudioFrame.h

@@ -57,6 +57,8 @@ class Q_AV_EXPORT AudioFrame : public Frame
      * then you can use clone().
      */
     AudioFrame clone() const;
+    AudioFrame mid(int pos, int len = -1) const;
+    void prepend(AudioFrame &other);
     AudioFormat format() const;
     void setSamplesPerChannel(int samples);
     // may change after resampling

src/QtAV/private/AVEncoder_p.h

@@ -71,6 +71,7 @@ class AudioEncoderPrivate : public AVEncoderPrivate
 public:
     AudioEncoderPrivate()
         : AVEncoderPrivate()
+        , frame_size(0)
     {
         bit_rate = 64000;
     }
@@ -79,6 +80,8 @@ class AudioEncoderPrivate : public AVEncoderPrivate
 
     AudioResampler *resampler;
     AudioFormat format, format_used;
+
+    int frame_size; // used if avctx->frame_size == 0
 };
 
 class Q_AV_PRIVATE_EXPORT VideoEncoderPrivate : public AVEncoderPrivate

src/codec/audio/AudioEncoder.cpp

@@ -80,4 +80,8 @@ const AudioFormat& AudioEncoder::audioFormat() const
     return d_func().format_used;
 }
 
+int AudioEncoder::frameSize() const
+{
+    return d_func().frame_size;
+}
 } //namespace QtAV

src/codec/audio/AudioEncoderFFmpeg.cpp

@@ -53,7 +53,6 @@ class AudioEncoderFFmpegPrivate Q_DECL_FINAL: public AudioEncoderPrivate
 public:
     AudioEncoderFFmpegPrivate()
         : AudioEncoderPrivate()
-        , frame_size(0)
     {
         avcodec_register_all();
         // NULL: codec-specific defaults won't be initialized, which may result in suboptimal default settings (this is important mainly for encoders, e.g. libx264).
@@ -62,7 +61,6 @@ class AudioEncoderFFmpegPrivate Q_DECL_FINAL: public AudioEncoderPrivate
     bool open() Q_DECL_OVERRIDE;
     bool close() Q_DECL_OVERRIDE;
 
-    int frame_size; // used if avctx->frame_size == 0
     QByteArray buffer;
 };
 

src/filter/EncodeFilter.cpp

@@ -34,7 +34,7 @@ namespace QtAV {
 class AudioEncodeFilterPrivate Q_DECL_FINAL : public AudioFilterPrivate
 {
 public:
-    AudioEncodeFilterPrivate() : enc(0), start_time(0), async(false), finishing(0) {}
+    AudioEncodeFilterPrivate() : enc(0), start_time(0), async(false), finishing(0), leftOverAudio() {}
     ~AudioEncodeFilterPrivate() {
         if (enc) {
             enc->close();
@@ -47,6 +47,7 @@ class AudioEncodeFilterPrivate Q_DECL_FINAL : public AudioFilterPrivate
     bool async;
     QAtomicInt finishing;
     QThread enc_thread;
+    AudioFrame leftOverAudio;
 };
 
 AudioEncodeFilter::AudioEncodeFilter(QObject *parent)
@@ -174,16 +175,35 @@ void AudioEncodeFilter::encode(const AudioFrame& frame)
     AudioFrame f(frame);
     if (f.format() != d.enc->audioFormat())
         f = f.to(d.enc->audioFormat());
-    if (!d.enc->encode(f)) {
-        if (f.timestamp() == std::numeric_limits<qreal>::max()) {
-            Q_EMIT finished();
-            d.finishing = 0;
+
+    if (d.leftOverAudio.isValid()) {
+        f.prepend(d.leftOverAudio);
+    }
+
+    int frameSizeEncoder = d.enc->frameSize() ? d.enc->frameSize() : f.samplesPerChannel();
+    int frameSize = f.samplesPerChannel();
+
+    QList<AudioFrame> audioFrames;
+    for (int i = 0; i < frameSize; i += frameSizeEncoder) {
+        if (frameSize - i >= frameSizeEncoder) {
+            audioFrames.append(f.mid(i, frameSizeEncoder));
+        } else {
+            d.leftOverAudio = f.mid(i);
         }
-        return;
     }
-    if (!d.enc->encoded().isValid())
-        return;
-    Q_EMIT frameEncoded(d.enc->encoded());
+
+    for (int i = 0; i < audioFrames.length(); i++) {
+        if (!d.enc->encode(audioFrames.at(i))) {
+            if (f.timestamp() == std::numeric_limits<qreal>::max()) {
+                Q_EMIT finished();
+                d.finishing = 0;
+            }
+            return;
+        }
+        if (!d.enc->encoded().isValid())
+            return;
+        Q_EMIT frameEncoded(d.enc->encoded());
+    }
 }