Just how much audio, in total, are you looking to store?
Example, stereo 44.1kHz sample rate 16 bits, is 5.94 seconds. As a continuous circular buffer, a net-size Flash chip will last a week, assuming you need rated retention time. But needing only a 6 second retention time, you'd probably get the better part of a year, maybe several years of useful life, call it planned obsolescence with a graceful failure mechanism (i.e., the audio gradually corrupts with pops and hiss -- which would be a good idea to detect and mute).
Huge Flash chips are readily available. You're literally paying as much, or more, for a small Flash chip, as for a gigabit one.
If this is for brief snippets, forget about it, stuff it in Flash. Even if it's continuous buffering, use Flash.
Otherwise, something like self refreshing DRAM (pseudo-SRAM). No idea if that's available in SPI flavor.
Your PDIP/SOIC requirement is easily satisfied: get a micro-SD breakout board. Now you have oodles of gigabytes on an SPI header. You'll never run out!
I don't get the lack of compression. Even on a $2 MCU, you have enough clock cycles to do at least a simple compression: mu-law scaling, Huffman encoding, various kinds of DCT, or fancier methods like SMASH or lzop or gzip. A combination of these will net you at least a very worthwhile 2x savings, or easily more than 16x if you allow some degradation in quality.
Tim