Yeah, did that. I gave audacity lame encoder 95-135 and it is awful, significantly worse than 128 constant which is also not ideal. Anything at all less than
144 constant or
110-150 variable
Sounds worse than q0 ogg vorbis on specific elements like cymbals clashes, drums, and sometimes even distorted / garbled vocals.
I also recently learned what opus was, and it’s even better than OGG, though up to 256 kbps per channel only.
Opus is so good, 12 kbps is actually listenable, with all lyrics and basic instruments of a song identifiable. It’s scratchy, and the volume and tone wavers, but still listenable.
A 12 kbps encoding in opus of a 7 minute song results in a file size of just 610 KB. For acceptable msuic quality though, 56 - 64 KBps is quite amazing, I didn’t know bitrate could be this low with such incredible sound quality.
Opusenc is probably installed on most Linux systems, encase the file paths in single ’ marks, and use the --bitrate setting.
It’s amazing that a 32 kbps opus file has clear, real sounding cymbal clashes, where as an mp3 need 144 constant to get close to that.
Even 110 - 150 variable mp3 doesn’t quite get it with cymbals, I guess the format thinks of it as unimportant, where as these other formats seem to have great respect for that sound quality.
To whoever asked Theora is used by some fansubs as an alternative to AAC or FLAC (rip coalgirls!). It’s not OGG files as it’s encapsulated in Matroska.
Opus’s general audio codec (CELT) is about 20 years newer (2011) than MP3 (1991), and still being improved for very low bitrates. Naturally, it’s better at lower bitrates! And more complex than MP3 - but negligible considering how much hardware and CPU power advanced between 1991-2011.
With sharp transients like cymbals, MP3 VBR really depends on the codec implementation to give it sufficient bitrate for that (short) block. LAME usually does a very good job at V2, but 110-150 kbps is V5-6 territory and not optimal.
Kinda the opposite, actually. MP3 has a special (“short”) window specifically to help deal with those (as @qqk already mentioned):
MP3 uses pre-echo detection routines, and VBR encoding, which allows it to temporarily increase the bitrate during difficult passages, in an attempt to reduce this effect. It is also able to switch between the normal 36 sample quantization window, and instead using 3× short 12 sample windows instead
And for things like percussion, applause, and other transients in general, you’ll get the best quality from temporal domain codecs like MPEG Layer II (MP3’s predecessor), Musepack (as I mentioned), DTS, or perhaps Dolby AC-3 (hybrid).