Try
www.winbond.com. They bought out ISD a while back. ISD make solid
state voice recorders. Some of these are switch driven, some require a
microcontroller to manage them.
If you only require playback the cheapest and simplest method would
probably be a small IIC or SPI flash memory, couple with a very cheap
micro, or some discrete logic to set start/end addresses.
Alternatively if you have any knowledge of microcontrollers you could
use a flash based part with large enough memory, and flash self
programming capability to implement this. An MSP430, for instance, using
straight forward 12 bit analog record and 12->8 bit log compression at
6800 samples/second (adequate for voice) would give very high quality
speech for around 8 seconds. While using something like the modified
CVSD algorithm posted on the Msp430 yahoo group ftp site with a sample
rate of 16kHz you would get very passable speech at just 2k/second, so
around 28 seconds in the MSP430F149.
1 second messages are not very long. Some years ago when I was
developing my own speech compression algorithms I studied average word
lengths, taking 20 different speakers, each reading 5 1000 word
passages. Without silence, which can be software inserted, I calculated
that the average spoken word required 0.4 seconds of storage.
Cheers
Al
