So I have some Plans to make further changes to the audio interface.
Basically:

1) Add an AUMODE_RECORD_ASYNC, which mirrors AUMODE_PLAY_ALL.  If the
   flag is not set, recording will be synchronous; i.e. the buffer
   will wrap around and silence will be inserted into the stream given
   to the user.  If the flag is set, recording will be asynchronous;
   i.e. an underrun will simply cause the recording process to be
   suspended until the data drains a bit.

2) Add a pause_{in,out}put interface to use the hardware pause feature
   if available.  In asynchronous modes, this will be used rather than
   inserting silence blocks for both pausing and for handling
   over/underruns.

The idea here is twofold: to introduce a way of toggling synchronous/
asynchronous recording, and to eliminate the use of silence blocks for
asynchronous modes.  For audio modes that cannot use silence blocks
(such as the MPEG modes on one of the Amiga chips), the set_params
routine will be expected to reject the encoding combined with a
synchronous timing mode.