Fri, 10 Dec 2004
Listening test for very low bit rate Ogg Vorbis encodings
The source is Democracy Now from July 20th 2004. It's the first two minutes of the show. It's a decent real-world test because there's music, speech from a studio mic, speech from a telephone and speech/music coming off tape, so you can A-B with a good spectrum of sound. For kicks I posted the times each encode took on my system.
System specs:
AMD Athlon XP 1600 @ 1400 Mhz
1000 MB DDR RAM
ATA 100 disk with aprox. 41 MB/sec throughput
Linux 2.4.26 Fedora Core 1
Original uncompressed recording (mono, 44100khz, 16 bit)
Ogg Quality 3
[lee@vorbis tmp]$ oggenc -q 3 dn2004-0720-listening_test.wav
File length: 2m 00.0s
Elapsed time: 0m 09.4s
Rate: 12.8233
Average bitrate: 73.9 kb/s
Ogg quality 0
[lee@vorbis tmp]$ oggenc -q 0 dn2004-0720-listening_test.wav
File length: 2m 00.0s
Elapsed time: 0m 08.3s
Rate: 14.4509
Average bitrate: 48.3 kb/s
Ogg quality 0, resampled to 22050khz
[lee@vorbis tmp]$ oggenc -q 0 --resample 22050 dn2004-0720-listening_test.wav
File length: 2m 00.0s
Elapsed time: 0m 05.1s
Rate: 23.4608
Average bitrate: 30.7 kb/s
Ogg quality 0, resampled to 11025khz
[lee@vorbis tmp]$ oggenc -q 0 --resample 11025 dn2004-0720-listening_test.wav
File length: 2m 00.0s
Elapsed time: 0m 03.6s
Rate: 33.2143
Average bitrate: 20.6 kb/s
Ogg quality -1
[lee@vorbis tmp]$ oggenc -q -1 dn2004-0720-listening_test.wav
File length: 2m 00.0s
Elapsed time: 0m 07.4s
Rate: 16.1914
Average bitrate: 37.1 kb/s
Ogg quality -1, resampled to 22050
[lee@vorbis tmp]$ oggenc -q -1 --resample 22050 dn2004-0720-listening_test.wav
File length: 2m 00.0s
Elapsed time: 0m 05.1s
Rate: 23.3804
Average bitrate: 23.6 kb/s
Ogg quality -1, resampled to 11025
[lee@vorbis tmp]$ oggenc -q -1 --resample 11025 dn2004-0720-listening_test.wav
File length: 2m 00.0s
Elapsed time: 0m 03.5s
Rate: 34.0317
Average bitrate: 16.4 kb/s
My conslusions are that compared to the original, the -1 qualities sound
just as acceptable as the 0 qualities. Since we are shooting for nothing more than intellegibility here, I'd say that the lowest possible quality is recommended for targeting a modem audience in a live stream.
The quality 3 sounds exactly the same as the original to my ears, so I'd say that's a good setting for targeting a broadband audience.
posted at: 01:42 | path:
/hacking/audio |
permanent link to this entry
