Just to add fuel to the fire: Vorbis is far more complex and advanced than ADPCM, so it can yield higher quality at the same bitrate. So lower bitrate compared to ADPCM isn’t a problem at all, depending on what the sound is trying to represent.
“White noise”-like sound compresses poorly, in my experience, probably due to lack of redundant information. ADPCM doesn’t do a good job at all of compressing it, it sounds really odd. I also tried MP3, and it isn’t good at it either, at least with a low bitrate. I haven’t tested this much, unfortunately. The only real advantage ADPCM has is that it’s really fast to decode, there’s not much to it at all, no transforms and such.
If you think about it, “telephone quality” uncompressed audio (8 KHz sampling rate and 8 bit samples, as used in the old PCM telephony systems, that were multiplexed into T1/E1 and such) is 64 kbps, considered “good enough” for voice (comparable quality to an analog telephone line). ADPCM would let you bring that up to 16 KHz (at less than 8 bits depth, in theory), so slightly less bad. However, even MP3 would be “usable” for music at 64 kbps. It’s not going to be any good, but it’s way better than the uncompressed sound would be at the same bitrate. Opus has reasonable quality at that same bitrate, apparently.
But I don’t think this really tells the whole story. Vorbis is VBR only. ADPCM is just companding, so the bitrate is constant (half of the uncompressed version). So you can’t just compare them by setting a “target” bitrate and seeing which one does a better job, I think. Vorbis is going to give very different results for its output bitrate, depending on the input and the quality setting.
I don’t know why compression gets such a bad rep, it’s such a useful tool. I mean, imagine if people tried to send uncompressed voice through, say, Iridium or a HF modem. 2400 bps doesn’t give you much room to work with at all, but with vocoders like Codec2 you can actually recognize that there’s a human talking on the other end!
But unless you have uncompressed samples or you are willing to create new ones, I think it’s a lost endeavor to re-encode those that we already have…