Increasing the number of samples per second (increasing the “bit rate”) can be useful for a few specific types of processing, for example if you need to align the phase of two separate recordings. It allows finer changes to be made in the time domain. If you think of digital audio as dots at regular time intervals on the analog waveform, then increasing the sample rate puts more dots (shorter time intervals) on that analog waveform. When recording, having more dots (samples) per second means that you can record higher frequencies. However, putting more dots onto a waveform that has already been digitally recorded makes no difference to the waveform - it just makes more dots on the waveform that is already recorded. Doubling the sample rate will add an extra dot between each existing dot on the recorded waveform, but makes no change other to that recorded waveform. Thus, when converted back to analog (on playback), the reconstructed analog waveform is exactly the same as it was before you added the extra dots. Using a very high sample rate rarely has any benefit, as even 44.1 kHz can accurately represent frequencies well over 18 kHz, which is much better than loudspeakers can perform, and better than most adults hearing.
Increasing the “bit depth” (“bit format”) is an increase in the number of bits per sample. This is very useful when processing audio because it allows the processing to be done more accurately, which is why Audacity uses 32 bit float by default. Processing audio changes the sample values in some way, for example, “amplification” multiplies each sample value by a specified amount. When processing in 16 bit format, the processed sample values are rounded to the nearest 16 bit value. This rounding is called “quantize error”. By increasing the number of bits per sample, quantize errors are reduced. For 32 bit float format, the quantize errors are so small that they are totally insignificant (totally inaudible), whereas 16 bit rounding errors may be just about audible (but very subtle because even 16 bit is a high quality format). If you are not processing, then increasing the bits per sample does nothing to the sound quality - it simply pads the sample values with additional zeros.
By analogy, say that you have three sample values: 23, 47, -16
If we were processing, say, reducing the volume by half (divide by two), then our original version would give rounded values because we are using whole numbers:
12, 24, -8
and if we then amplified to double the volume (multiply by 2), then we can see the errors:
24, 48, -17.
Now lets do the same again, but first we will increase the resolution:
We can increase the precision by writing them as: 23.000, 47.000, -16.000
Note that the values are exactly the same as the original sample values, so we have added nothing to the “quality”.
Then when we divide by 2 we get:
11.500, 23.500, -8.000 (working to three decimal places)
Then when we multiply by 2:
23.000, 47.000, -16.000
We have retained the full fidelity, and this is the benefit of processing in a high sample format.
Does that help?
)