Normalize Vs Volume increase Options

[tinpanalley]

I'm having trouble figuring out which of the two is better for me to create a vinyl recording at the highest volume without clipping.

I capture a record via my preamp into Audacity, and I get what seem to be relatively low levels even though I make sure the levels are right when I record (Peak: -6.4db, RMS -24.6db). Am I doing something wrong? Or are those levels right? These recordings don't seem as loud as mastered CDs.

Anyway, to get the most accurate sound possible, should I use normalize or just increase the volume? If normalizing makes everything sound the same at a peak level, doesn't that completely screw with the dynamic range of the music?

Thanks!

[AndyH-ha]

There exist a continuously varying analogue signal from out of the analogue part of the recording chain. It is sampled by the ADC (in your case, inside the USB device). The bit depth is the precision to which it is sampled.

2 to the power of the number of bits is the number of different voltage values that can be output from the DAC (i.e. 2^16=65536 steps between minimum negative voltage and maximum positive voltage -- half of these for greater that 0, half for less than zero, minus one or two for the sign).

The ADC must “choose,” from these 65536 possible values, the one that is closest to the “actual” value of the signal voltage. The more bits it can use (i.e. 24 bits), the smaller each step, so the closer the choice will be to the real analogue voltage (as measured with infinite precision, or at least more precision than the ADC can manage).

24 bits is the limit of the hardware. However, no soundcard can actually do 24 bits. Thermal noise fills the lowest order bits due to the nature of not living in a extra cold part of the universe, so voltages very (very) near zero are lost in noise.

16,777,216 steps for 24 bit covers exactly the same range as 65,536 steps for 16 bit, so the quantization error (difference between “actual” voltage and what the ADC can record) is that much smaller (2^24/2^16). If fact it isn’t possible to hear the difference but when you do post recording processing, the same kind of errors occur in each calculation, so they add up.

There are a variety of 32 bit floating point formats but they are basically a 24 bit mantissa and a 8 bit exponent. This lets the data scale as necessary. Recording is not different between 24 bit and floating point but calculations errors are smaller and it is almost impossible to clip (unless you forget about >0dB values when you convert to integer).

Any format, 16 bit integer, 24 bit integer, floating point (32 and 64 bit varieties) works fine for LP recordings, which have so much intrinsic noise that they obscure the differences. Still, why not do as well as you can?