HOWTO Create an SVCD with MPEG Multichannel Sound

Software Requirements

For the transcoding process you need the following software:

The 0.9 branch of the Linux DVB driver for the pes2vid_es program (newer versions from the CVS repository of linuxtv.org might also work)
The MPEG decoder mpeg2dec
The MJPEG tools, especially the MPEG encoder mpeg2enc
The AC3 decoder ac3dec with my 6-channel patch (also contains the extract_ac3 program to extract AC3 streams from MPEG system streams)
The dist10 MPEG1 layer 2 audio encoder
The mpeg2_movie package for its mpeg3cat/mpeg3split utilities
sox, a commandline sound processing program for UNIX (included in most Linux distributions)

You can find information about these packages in my Software section.

The Transcoding Process

The transcoding process is the more or less the same as described in HOWTO create SVCDs from DVB recordings. The only difference is in the handling of the audio stream, since we now have AC3 audio as the input:

```
for i in 1 2 3 4 5 ; do \
  (extract_ac3 $i.vdr -s 1 > $i\_de.ac3);
done
```
Extract the Dolby Digital (AC3) streams from the MPEG program streams (VDR files) and write to files 1_de.ac3 ... 5_de.ac3.
```
for i in 1 2 3 4 5 ; do \
  export AC3_6CH=$i\_en; \
  ~/src/ac3dec-010404-6chout/ac3dec -o null $i\_de.ac3; \
done
```
Take AC3 files, decode them, and write the PCM files. For each AC3 file, e.g. 1_de.ac3, six files 1_de_xx.ac3 are written where x is c (center), l (front left), lfe (low-frequency effect), ls (left surround), r (front right), rs (right surround), respectively.
Note that you need my patched version of ac3dec to obtain the six-channel write feature.
```
for i in *.pcm; do \
  sox -t raw -r 48000 -s -w -c 1 $i -t raw -r 44100 -s -w -c 1 $i.44; \
done
```
This step resamples each PCM file from 48000 Hz to 44100 Hz sampling rate (needed for compliant SVCDs).
```
rm *.pcm
for i in *.44; do mv $i ${i%.44}; done
```
This deletes all 48000 Hz PCM files and renames the 44100 Hz files.
```
~/src/dist10/tool/pcm2aiff/pcm2aiff2 "1_de" -c5 -r0
```
This line takes the 1_de_c.pcm ... 1_de_rs.pcm files as input and combines them into a single AIFF file suitable for the MPEG encoder.
```
export MPEGTABLES=~/src/dist10/mc/encoder/tables
~/src/dist10/mc/encoder/musicin -b 224 -m s -n d -o -L -e 1_de.aiff
```
This encodes the AIFF file into a MPEG layer 2 multichannel 3/2.1 stream. You should not go below a bitrate of 224 kbit/s.

You can use my transcode_mc.sh script to automate this tedious process. Note that there are some configurable parameters at the beginning of this script:

ac3_parts Add a number for each MPEG stream file with AC3 included (must be of the form n.vdr).
pcm44_parts The number of bytes for each PCM file. This must correspond in playback time to the size of the video MPEG file which is later to be multiplexed with the resulting MPEG audio multichannel file. Just take the number of frames of the transcoded video, divide by 25 (to get number of seconds with PAL video), and multiply with 2*44100 (44100 Hz sampling rate, 2 bytes per sample) to get the size of the intermediate PCMs.
svcd_parts Add a number each svcd MPEG file you want to create.
lang I use to letter codes for the language of the AC3 stream to transcode, just a file name convention.
ac3_id AC3 id of the AC3 stream to extract and transcode.

Note: You also need my split_pcm program for this script.

The resulting MPEG audio stream can then be multiplexed as described in HOWTO create SVCDs from DVB recordings. Since MPEG multichannel audio is backwards-compatible to regular MPEG layer 2 audio the SVCD should play back on every player that can play SVCDs. My Onkyo receiver recognizes the audio as being MPEG 3/2.1 when it comes via S/P-DIF.

Variations Without VDR

If you have gotten your MPEG program streams from some other source and not from VDR (or you don't like to modify and restart VDR every time), you can use the mpeg3split and mpeg3cat programs from mpeg2_movie/libmpeg3 to first concatenate your elementary video or audio streams and then split them again (you need my patched mpeg3split for files larger than 2 GB). The steps would be for the video part:

```
for i in 1 2 3 4 5; do \
  (~/src/dvb-cvs/DVB-0.8/apps/mpegtools/pes2vid_es video_ps_$i.mpg > $i.mpg); \
done
```
This takes the files video_ps_1.mpg ... video_ps_2.mpg and extracts the elementary video streams into separate files. Read mpegtools/README for information on how to obtain pes2vid_es.
```
mpeg3cat 1.mpg 2.mpg  ... > es_video.mpg
```
This concatenates the elementary video streams into one file. The resulting file may be larger than 2 GB, so make sure that your filesystem supports this (ext2 with a Linux 2.4.1 kernel works for me).
```
~/src/mpeg2_movie-1.5/libmpeg3/mpeg3split -b 1308957000 es_video.mpg
```
Use mpeg3split to create evenly sized elementary video streams (you need my patched version with large file support). These can then be transcoded to SVCD format. The transcoding process (i.e. the mplex program) gives you information on the number of video frames in each elementary video stream. You should record this information as it is needed to properly cut the audio streams.

Now we come to splitting the audio for MPEG multichannel encoding. You should already have created the PCM files with 44100 Hz sampling rate. Now follow these steps:

cat 1_de_l.pcm ... 5_de_l.pcm > de_l.pcm
...
cat 1_de_rs.pcm ... 5_de_rs.pcm > de_rs.pcm

Combine individual PCM files into 6 PCM files for each channel.

The de_x.pcm files need to be split again so that they match the elementary video streams in playback duration. Suppose the first video stream has 55600 frames, then each 1_de_xx.pcm file should have the size 55600/25*2*44100 = 196156800 bytes. Explanation: 55600/25 is the playback duration of the video in seconds. Each second has 44100 samples of audio with 2 bytes length. Use UNIX tools like head and tail to split the PCM files.

Now you can create the AIFF file and encode it to MPEG multichannel.

After the final multiplex step you should be especially careful to check synchronization of audio and video (e.g. with xine, before writing to CD-R). If there is a problem you can re-multiplex with a different delay by giving the -v xxx or -a yyy options to mpegtools/mplex where xxx and yyy are the video and audio delays, respectively.

© 2001 Peter Hofmann
Last Update: 01.07.2001