[vorbis] Performance of low quality / low sample rate

[Shawn Riley]

Hi everyone.

I did a rough recording of an instrumental (electric piano sound) & e-mailed it to a friend in Vorbis 11025 Hz / mono. I was seeking a bitrate in the range 8-16k/sec. The song is 2:55 in length. q=-1 happily achieved a 12.6k/sec bitrate. All file sizes I mention in this are for files without informational tags.
And I hope this isn't interpreted as trying to plug my music or website (which probably isn't worth plugging anyway. ;-)

Afterwards, I was curious to compare Vorbis's performance with MP3, I used Lame 3.93.1 from http://mitiok.cjb.net/ & Oggenc 1.0 available from Vorbis.com
Basically, the lowest quality Lame stream was 289900 bytes (13265 bit per second), & it sounded better in quality than q=-0.93 Vorbis which resulted in the same file size. So I increased the quality until they sounded approximately the same quality across the whole song. I ended up using Vorbis at q=1.2 which results in 434837 bytes (19897 bit per sec).
The instrument sound is very tonal, & the sharp "ping" at the start of each note which characterises the sound, is also tonal according to Cooledit Pro's spectral view (512 band), & I wonder if at q=-1, Oggenc was consistently using blocks that are too long for the note-on sound, then encoding many frequencies around it, thus increaing the file size. It's a thought I came across because in CEP, the frequency analyses showed a "softer" spectrum for the Vorbis file. Also during the long extended notes, Oggenc seems to be flicking between frequency bands.
But Vorbis had comparable pre-echo to MP3 at comparable bitrates, so I might just be talking through my hat for this possibility.
Since Vorbis does pretty well in comparison to Lame at this bitrate in other genres, I think it might be that Vorbis doesn't like this type of instrument, especially in solo.

Anyway, the full song will be / is, up at-
 http://www.angelfire.com/rock3/roleypup/inst.tar
Apologies for the huge file size (1026KB). I wanted it to be as close to lossless as it could get.
Encoded with Oggenc 1.0

Shawn Riley

PS- More info
Encoding command strings below-
"oggenc -q-0.93 --downmix"
 and
"lame -a -q0 -v -V9 -B32 --lowpass 4980 --lowpass_width 0"
produced the same size.
For those unfamiliar with Lame, it's downmixing, using slowest/best compression, selecting VBR q9 level (lowest qual), limiting frame size to 32k/sec frames, & setting lowpass in Hz. Lame was more vicious in its lowpass filtering at this bitrate, so I had to manually adjust it to approximately the same cutoff as what Oggenc is using.

Decoding command strings-
"oggdec"
and
"lame --decode"

Keyboard model - Korg SP-200
Patch "E.PIANO 2", Bank 1.
No Reverb
Chorus Depth "medium"
Everything else default.

Range of notes in song-
Lowest - 1st C# below middle C (138.5 Hz)
Highest - 3rd G# above middle C (1661 Hz)
--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.