The bit depth issue
It’s hard to hear the difference between 24-bit and 16-bit audio during music, especially when you don’t know what to listen for. In mastering it’s really important to know your problems and how they actually sound. The main difference is the lower headroom and the higher noise floor (in the 16-bit version). To this day I haven’t found any material that present this difference in a good audible way. That is why I decided to attempt to do this.
Understanding bit depth
16-bit and 24-bit PCM (pulse code modulation) is a digital way to present analog sound with evenly spaced samples. These samples represent amplitude and are quantized to a digital step of the nearest possible value.
16-bit digital has a maximum SNR (signal to noise ratio) of 96 dB and 24-bit digital audio has 144 dB, though the AD-conversion only can handle 123 dB (equivalent to 22-bit). It’s important that the process of making a bit reduction (if necessary) is the last thing you do in the mastering. Otherwise you can accidentally accentuate the noise created by the bit reduction.
The difference between 16-bit and 24-bit audio
When just listening to music this difference is hard to notice but if you turn up the volume during a fade out of a song (16-bit) you can actually hear the noise floor sooner than in a 24-bit depth sound.
To recreate this I have lowered the audio of a drum beat and then faded it out slowly, I then bounced it to 16-bit and imported it to a new project. After that I turned up the gain and bounced it again at a 24-bit depth resolution. I did this first without dither, then with dither and finally with dither with noise shaping.
24-bit depth resolution
In this case the span of the signal to noise ratio is very large and the volume of the audio looks relevant.
At the very end of the fade you can actually hear a little bit of noise even here. Remember that this audio is originally at very low volume.
16-bit depth resolution
After the audio has been gated the transients resurface above the threshold and creates a noisy attack that resurface the sound that slowly decays and finally gets gated again.
Where the original 24Bit audio faded smoothly between volumes this one doesn't have the same signal to noise ratio and therefore clips the low volume audio. This is called quantization distortion. Remember that this audio is originally at very low volume.
16-bit depth resolution with dither
This is also a 16 bit depth audio but with dither enabled.
Dithering is the process of mathematically removing the undesired distortion and replace it with a constant noise level. You can actually hear the noise being present in the whole clip and at the end you get less of that sudden ”noisy threshold” though you still hear the noise when the transients break through. The drums in themselves doesn’t sound as gated as in the previous example. The noise is very present but not that aggressive sounding but still very easy to hear. Remember that this audio is originally at very low volume.
16-bit depth with dither
16-bit depth resolution with dither & noise shaping
This is a 16 bit depth audio with dither and noise-shaping enabled. After the dithering has removed the undesired distortion and replaced it with a constant noise level the noise shaping is then altering the spectral shape of the errors.
This means that it’s moving the noise to a less audible (and more desirable frequency). This is why there is so much red (high volume sound) in the high frequencies on the spectrogram.
Other than that this audio sounds and looks way more desireable than the other 16Bit audio versions. It looks and sound more like the original. The noise in the 24Bit version is even more preferreable (to me). Remember that this audio is originally at very low volume.
16-bit depth with dither and noise shaping
Visual bit reduction
The colors in these visuals demonstrate loudness in the frequency spectrum (from green to red).
- You can clearly see that the 24Bit audio looks very smooth.
- In the 16Bit audio (without dithering or noise shaping) you can see the threshold gate the audio into silence it looks glitchy and weird.
- The spectrogram in the dithered version (without noise shaping) looks flat in loudness and even in frequency. That is because the noise is beeing present (and loud) all over the frequency spectrum.
- Finally the 16Bit audio with dithering and noise shaping doesn't look as flat but the noise looks like it has moved in frequency and become alot louder. As if the noise from previous version has moved from all over the spectrum to the high frequencies.
Quantization errors with noise only versions
I decided to isolate the noise from the quantization errors (also known as truncation) to present the (almost) pure noise from the conversion. I did this by playing the original file and the bit reduction sound simultaneously and phase reverse the original. This way the sound will only present the difference between the two. In the un-dithered version there is actually some of the original sound present, this is due to the gated effect of the reduced signal to noise ratio.
16Bit quantization error noise
Some of the original sound present in this sound is due to the gated effect of the reduced signal to noise ratio.
16Bit with dithering quantization error noise
16Bit with dithering and noise shaping quantization error noise
If you have problems playing the audio here you can listen on soundcloud. But remember that there are some really nasty conversion artefacts on soundcloud.