Modify the Ogg/Speex demuxer and the libspeex decoder so that they always treat

a packet of Speex frames as a single frame.

Originally committed as revision 19734 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/libspeexdec.c

@@ -53,6 +53,8 @@ static av_cold int libspeex_decode_init(AVCodecContext *avctx)
         avctx->sample_rate = s->header->rate;
         avctx->channels    = s->header->nb_channels;
         avctx->frame_size  = s->header->frame_size;
+        if (s->header->frames_per_packet)
+            avctx->frame_size *= s->header->frames_per_packet;
 
         mode = speex_lib_get_mode(s->header->mode);
         if (!mode) {
@@ -98,7 +100,7 @@ static int libspeex_decode_frame(AVCodecContext *avctx,
     int16_t *output = data, *end;
     int i, num_samples;
 
-    num_samples = avctx->frame_size * avctx->channels;
+    num_samples = s->header->frame_size * avctx->channels;
     end = output + *data_size/2;
 
     speex_bits_read_from(&s->bits, buf, buf_size);
@@ -113,12 +115,13 @@ static int libspeex_decode_frame(AVCodecContext *avctx,
             break;
 
         if (avctx->channels == 2)
-            speex_decode_stereo_int(output, avctx->frame_size, &s->stereo);
+            speex_decode_stereo_int(output, s->header->frame_size, &s->stereo);
 
         output += num_samples;
     }
 
-    *data_size = i * avctx->channels * avctx->frame_size * 2;
+    avctx->frame_size = s->header->frame_size * i;
+    *data_size = avctx->channels * avctx->frame_size * sizeof(*output);
     return buf_size;
 }
 

libavformat/oggparsespeex.c

@@ -40,12 +40,22 @@ static int speex_header(AVFormatContext *s, int idx) {
         return 0;
 
     if (os->seq == 0) {
+        int frames_per_packet;
         st->codec->codec_type = CODEC_TYPE_AUDIO;
         st->codec->codec_id = CODEC_ID_SPEEX;
 
         st->codec->sample_rate = AV_RL32(p + 36);
         st->codec->channels = AV_RL32(p + 48);
+
+        /* We treat the whole Speex packet as a single frame everywhere Speex
+           is handled in FFmpeg.  This avoids the complexities of splitting
+           and joining individual Speex frames, which are not always
+           byte-aligned. */
         st->codec->frame_size = AV_RL32(p + 56);
+        frames_per_packet     = AV_RL32(p + 64);
+        if (frames_per_packet)
+            st->codec->frame_size *= frames_per_packet;
+
         st->codec->extradata_size = os->psize;
         st->codec->extradata = av_malloc(st->codec->extradata_size
                                          + FF_INPUT_BUFFER_PADDING_SIZE);