F – each sample stored as a floating point number
S – each sample stored as signed integer
U – each sample stored as unsigned integer
16 – each sample takes 16 bits
32 – each sample takes 32 bits
BE – each sample stored in big-endian format (endianness)
LE – each sample stored in little-endian format
The reason this matters is due to needing to preserve the audio that is above and below the range of human hearing from the extraction source.
If you're talking about frequencies above the range of human hearing then sampling frequency is the most important parameter for you. Take the highest frequency you want to preserve, drop everything above (see low-pass filter) and sample with twice the frequency. This procedure prevents aliasing.
Preserving of low frequency sounds ("below the range of human hearing") in theory comes easy because even poor sampling frequency will suffice.
Your hardware may have trouble registering or playing extreme frequencies though.
If "above and below the range of human hearing" is about loudness and quietness then you will need high dynamic range. In such case you may like floating point better than integer.
AFAIK signed vs unsigned and big-endian vs little-endian don't affect the quality of digital audio; i.e. you can convert from signed to unsigned without data loss, the same between big-endian and little-endian. These settings need to match between encoder and decoder.
Details of audio processing, of analog-to-digital conversion – are out of scope of Super User. Use the links I gave you to boost your research.