I've played around a bit with Speex, the free speech codec that can be embedded into Ogg packets. Thanks to Speexdrop, a very useful tiny GUI for Speex encoding, I've been able to encode an interview I've done recently into Speex.

There are two voices recorded in the file: the one of the person interviewed (female), and mine (male). The file is 1 minute 47 seconds long. I've set the complexity to 10, 1 frame is stored in each Ogg packet, the stereo sound has been downmixed to mono, VBR is on, and samplerate is 44100 Hz.

Here are the results:

Quality File size Average bitrate OK for broadcast?
0 173 kB 12.93 kbps no
1 196 kB 14.65 kbps no
2 222 kB 16.60 kbps no
3 257 kB 19.21 kbps nearly
4 306 kB 22.88 kbps nearly
5 348 kB 26.02 kbps nearly
6 377 kB 28.19 kbps nearly
7 445 kB 33.27 kbps nearly
8 528 kB 39.48 kbps yes
9 658 kB 49.20 kbps yes
10 684 kB 51.14 kbps yes

To my ears, being OK for broadcast means "better than a classic phone line".