To start recording voice, you only have to turn it on using the play button, then press the record button to start. To stop, press the record button again. Each time you toggle the record button, a new voice file is created. In other words, there is no pause button.
The recording format is ADPCM (adaptive differential pulse code modulation). According to sndfile-info (a Linux tool to examine sound files), the format is IMA-ADPCM. There are variants of .WAV files, and this variant is not the most compact one. It takes 8055 bytes per second in the voice-quality setting. This translates to 29MB per hour.
In Linux, I can use sndfile-convert to convert the format to a more compact encoding called GSM610 that is less than one half the size of IMA-ADPCM. GSM610 at the same sampling rate (16KHz, 8-bit) uses 3250 bytes per second, or about 12MB per hour. Note that there is no loss of quality going from IMA-ADPCM to GSM610. By the way, Windows (XP at least, probably the same with most other versions) knows how to play GSM610 encoded wave files.
Lectures and presentations can be recorded just fine at 16kHz sampling rate. For music, you can select 44.1KHz instead. Since the recorder can only record one channel, I do not intend to record any music anyway. It is nice to know, however, that I have an option to record at a higher sampling rate.