It depends how accurately you want to reproduce the voice. The maximum frequency that can be reproduced from a digital signal is one-half the sampling rate. This is also known as the Nyquist frequency. However, because an analog signal needs to be lowpass-filtered before being digitized to prevent aliasing, the sampling rate must be higher than twice the Nyquist frequency to take into account the filter characteristics. A filter with a very high a rolloff rate can allow the use of a sampling rate about 2.5 times that of the highest frequency of interest. Better results are obtained using a filter with a lower rolloff rate and a higher sampling rate, typically 4 or 5 times the Nyquist frequency. The human voice has a bandwidth of about 8 to 9 kHz, which would require a sampling rate of 16 to 18 kHz to reproduce, assuming the use of a perfect lowpass filter. The spoken voice has a lower bandwidth, about 4 kHz, which would require a sampling rate of 8 kHz or higher to reproduce. Low sampling rates may not reproduce a voice properly, with the result that it may sound as if the person is talking on the telephone. The standard sampling rate for CDs is 44.1 kHz, which barely accommodates the 20 kHz range of the human ear. This is a good place to start for home recordings, particularly if the material is to be recorded on a CD. The industry standard for digital tape recordings has been 48 kHz for some time, although this is slowly being replaced by higher sampling rates (88 kHz and higher). Recordings should be done at a minimum resolution of 16 bits, which is the bit depth used with audio CDs. I would suggest starting with a sampling rate of 22.05 kHz, half the standard CD rate, and compare the results with a recording done at 44.1 kHz. If it sounds substantially the same, you could use the lower rate. If the recordings will ever end up on a CD, I would strongly advise using the standard CD sampling rate of 44.1 kHz, which is universally supported on optical disk players.