After some frustration trying to get mp3splt to work, I caved in and wrote a script to split apart a large audio file into many 3min chunks. Save this in mp3split.sh:
#!/bin/bash
big="$1"
duration_stamp=$(ffmpeg -i "$big" 2>&1 | grep Duration | sed 's/^.*Duration: *\([^ ,]*\),.*/\1/g')
title=$(ffmpeg -i "$big" 2>&1 | grep "title *:" | sed 's/^.*title *: *\(.*\)/\1/g')
# get minutes as a raw integer number (rounded up)
prefix=$(basename "$big" .mp3)
echo $duration_stamp
mins=$(echo "$duration_stamp" | sed 's/\([0-9]*\):\([0-9]*\):\([0-9]*\)\.\([0-9]*\)/\1*60+\2+\3\/60+\4\/60\/100/g' | bc -l | python -c "import math; print int(math.ceil(float(raw_input())))")
ss="0"
count="1"
total_count=$(echo "$mins/3+1" | bc)
while [ "$ss" -lt "$mins" ]
do
zcount=$(printf "%05d" $count)
ss_hours=$(echo "$ss/60" | bc)
ss_mins=$(echo "$ss%60" | bc)
ss_stamp=$(printf "%02d:%02d:00" $ss_hours $ss_mins)
ffmpeg -i "$big" -acodec copy -t 00:03:00 -ss $ss_stamp -metadata track="$count/$total_count" -metadata title="$title $zcount" "$prefix-$zcount.mp3"
ss=$[$ss+3]
count=$[$count+1]
done
The execute mp3split.sh my-long-file.mp3. This will output a sequence of files:
my-long-file-00001.mp3
my-long-file-00002.mp3
my-long-file-00003.mp3
my-long-file-00004.mp3
...
Each will retain the meta data from the original file except the file number will be appended to the track name and the track number will be set accordingly (i.e. this will work well for splitting enormous audiobook files into file lists that play in the correct sequence on an iphone).
Note: mp3splt really seems like the right tool for this. It supposedly has fancy features like silence detection and presumably won't reload the file for each new split.