I've used Winbond single-chip voice recorders to good effect in the past, for example in this project. I recently had a need to incorporate pre-recorded audio in a project so I revisited this chip line. There were some changes. First, Winbond has spun out its speech recorded chip business as a company called Nuvoton. Second, the sweet spot seems to be a new line of chips, the ISD1700 series, where SS is the number of seconds that can be recorded - ranging up to ISD17240. Third, to promote these chips they've released a line of development kits with the chips mounted on a small circuit board (ISD-COB1700) and with all the fidgety analog stuff already wired up - at a cost only slightly more than the chips themselves. You can program these chips fromusing play/record/stop switches, or via SPI. They can drive a speaker directly, and receive input either from an electret microphone (built into the ISD-COB1700), or from line input - e.g. from a laptop, an mp3 player, and so on. Below is an ISD-COB17240 wired up to get some record from my netbook.
In spite of developing and selling at low cost such a convenient development kit, Nuvoton seems to want to discourage any experimention. For one, the set of SPI commands is not part of the ISD1700 documentation. Instead, one gets referred to the "ISD1700 Series Design Guide" - which is not available on the Nuvoton web site. Fortunately some web search brings it up, e.g. here. For another, one cannot record from the built-in microphone when controlling the ISD chip using SPI. I had intended to integrate programmatic voice recording as part of my project and spent a day of frustration trying to make it work only to be greeted by silence. I could make the line-in recording work, so I started to devise plans to bypass or replace the microphone-in circuit, when I bothered to examine that part of the ISD-COB1700 schematic. A part of the microphone circuit which should be connected to ground is instead connected to the not-REC input of the ISD1700, and is grounded only when the REC button is pressed (this schematic is repeated several times in the various ISD1700 documents). The solution is simple - bypass this part of the circuit as illustrated in the figure below, cutting the connection to not-REC and adding a connection to ground. I was dumbfounded when I figured this one out - its like finding a bug in the compiler.
The documentation is surprisingly well written in light of stupid problems I encountered in using this chip. However there are a couple of tricky issues which aren't well documented: