DSP resample uses 32-bit float?

[Someone7272]

Mainstream/top 40 record labels are finally starting to release music in 24-bit 48 KHz. But my car (iPod USB protocol) can't play 48 KHz, it has to be 44.1 KHz, doesn't seem to matter if it's 24-bit or 16-bit. That means I have to resample some my music. Resampling 48 KHz to 44.1 KHz will likely result in clipping, if the mastered track is already approaching -0dBFS.

If I use the DSP to resample, the audio will likely be clipped by the resample stage. But if I also use a second DSP step for normalizing to -0.1dBFS, will this preserve audio data, that would be clipped if I didn't? Note that I'm working with lossless FLAC source files, the output will be resampled 44.1 KHz ALAC.

[Spoon]

Use the dsp effect bit depth and set to float before the resample

[Someone7272]

Use the dsp effect bit depth and set to float before the resample

Does this apply even if I am using the same bit depth? For example, 24-bit 48 KHz resample to 24-bit 44.1 KHz?

I wanted to avoid messing with bit depth, cause then dithering comes into effect, but I’m not pro enough to understand how it works or what it does. Other than it seems to reduce audio quality (found this out the hard way with Audacity’s default global dithering).

If I set the bit depth to float as a DSP effect, can I avoid having to dither if the source and output is both 24-bit?

Otherwise, to avoid dithering, I’m considering:

1. DSP normalize -3 dB for resample headroom
2. DSP resample 44.1 KHz
3. DSP normalize -0.1 dB for line level output

If this is a bad idea or violates best practice, please explain why?

[Spoon]

Doing any maths will change the bit depth internally, Resampling cannot be done at 24 bit, it has to be floating point (internally it is 64 bit double)

So:

Bit depth (32 bit float) <<< which can hold 24 bit without losing quality
Resample
Volume Normalize (peak to peak or EBUR128) to ensure no clipping.
Bit depth 24 bit, no need for dither at 24 bit.