In this lesson, we’ll break down bit depth. Bit depth is a key component of digital audio. You’ve probably noticed a list of available bit depths when ever you’ve created a new session or bounced your music. In this lesson, we’ll break down exactly what bit depth is. We’ll also look at the key bit depths including 16 bit, 24 bit and 32 bit float to discover which bit depth to use and when.
Before we go any further, if you want to make better music in your home studio, then be sure to download your free home studio bundle here (EQ cheat sheet, compression cheat sheet, & vocal guide).
What is bit depth in audio
In order to record sound digitally, your analogue to digital convertor captures thousands of samples of the analogue signal every second. Each sample is assigned a value which is used to represent the amplitude of that sample. Bit depth defines the number of values that are available in each sample. As a result, bit depth defines the dynamic range of digital audio. Let’s look at how it does this…
When your analogue signal is sampled, its amplitude is matched to the closest of the available digital values. This process is called quantization. However, no matter how many values are available, the amplitude of the analogue signal will not always correspond perfectly with an available value. Sometimes the amplitude of the signal will fall between two values.
When the amplitude of the analogue signal does not line up perfectly with an available value, the analogue signal’s amplitude has to be rounded to the closest available value. This is referred to as quantization error. These quantization errors introduce quantization noise into the signal. Quantization noise is heard as low level white noise. We refer to the level of the quantization noise as the noise floor. The term noise floor refers to the signal which exists in an audio system as a result of the combination of any and all inherent noise or unwanted sound.
An important thing to note is that the higher the bit depth, the lower the level of the noise floor. As such, a higher bit depth provides greater dynamic range. Dynamic range refers to the difference between the loudest and quietest sounds that can be captured and reproduced. At the upper limit of the dynamic range is 0dBFS, above which any signal will clip. At the lower limit of the dynamic range is the noise floor. Any signal quieter than the noise floor will be masked by the noise floor.
16 bit vs 24 bit
The most common bit depths for recording and bouncing audio are 16 bit and 24 bit. 16 bit provides each sample with 65,536 possible amplitude values. 24 bit provides each sample with 16,777,216 possible amplitude values. As such, 16 bit provides you with 96dB of dynamic range between the noise floor and 0dBFS. 24 bit provides you with 144dB of dynamic range between the noise floor and 0dBFS.
How does bit depth affect sound quality?
There is something of a misconception that higher bit depths result in better quality audio or higher fidelity sound. A question that people ask is “does 24 bit sound better than 16 bit”? This confusion likely originates from the fact that a higher bit depth is referred to as having a higher ‘resolution’. This makes people think of the resolution of digital images in which the higher the resolution, the clearer and more detailed the picture. But that’s not the case here.
Here, the resolution that we are talking about is information resolution. So a 24 bit audio file won’t sound more detailed and clear than a 16 bit audio file. Instead, as we’ve already established, it simply means that a higher bit depth will have a lower noise floor. In fact, even when converting from a high bit depth to a lower one, provided you use dither, the only difference should be that the lower bit depth version has a higher noise floor. That’s because dither will rectify the occurrence of truncation distortion when converting to a lower bit depth.
What bit depth should I use for recording?
When it comes to recording, even at 16 bit, the noise floor is very low. So you could record at 16 bit and still have a decent amount of dynamic range. A more common practice however, and something of a standard, is to record at 24 bit. Simply because recording at 24 bit gives you more dynamic range to work with. It offers you the ability to record at more conservative levels to avoid going anywhere near the level of clipping whilst still being well above the noise floor.
Recently, a few interfaces have come to market which offer 32 bit recording. Note that I’m talking here about 32 bit fixed point, not 32 bit float, which we’ll move onto in a moment. 32 bit offers 4,294,967,296 values and a dynamic range of 192dB. But do we really need this much dynamic range? At 24 bit, you already have a dynamic range which is greater than the dynamic range of the human ear. In the human auditory system, the threshold of hearing is 0dB SPL, whilst the threshold of pain is around 120dB SPL. As such, the 144dB of dynamic range that 24 bit offers is already high enough to capture sounds lower than we can hear and louder than we can tolerate. So the 192dB of dynamic range offered by 32 bit fixed point is completely unnecessary.
What bit depth should I use to bounce my song?
So, what bit depth should you use for the final bounce of your song? That’s the audio file that you will upload to a distributor to release on iTunes, Spotify etc. Well, the standard bit depth for this for a long time was 16 bit. That was because to burn an audio file to an audio CD, the file has to be 16 bit. But many online platforms now allow you to upload files at 24 bit. But is this really necessary? Well, even if you were working with really dynamic material like classical music for example, it would still only use about 50dB – 60dB of dynamic range at the most. The majority of other genres will use much less. As such, the 96dB of dynamic range of 16 bit is more than adequate to represent the dynamic range of your music perfectly.
What is 32 bit floating point in audio?
With the basics covered, let’s move onto 32 bit floating point. You’ve probably seen this as one of the options alongside 16 bit and 24 bit when you create a new session or bounce audio.
First, it’s important to know that bit depths come in 2 different forms. The first is ‘fixed point’, also known as ‘integer’. The second is ‘floating point’. So far we’ve talked about 16 bit, 24 bit and 32 bit fixed point bit depths. Unlike fixed point bit depths, floating point bit depths are used only for internal processing in a DAW. They’re not suitable for use as the bit depth of your final track which you will then upload to a digital distributor. That bounce needs to have a fixed point bit depth. So you would really only bounce down to 32 bit float if the audio file was going to be imported into a DAW.
Unlike fixed point bit depths which have a ceiling at 0dBFS and a noise floor a certain distance beneath, 32 bit float uses its dynamic range differently. 32 bit float has a huge dynamic range of 1528dB, but instead of having a ceiling at 0dBFS, it has the upper limit of its dynamic range 770dB above 0dBFS and its lower limit 758dB beneath 0dBFS. As such, this system provides you with the ability to push the levels of things like your instrument channels and the input/output of plugins up and down within your DAW without having to worry about either falling below the noise floor or clipping.
32 bit float vs 24 bit
So, is it better to work with 32 bit floating point files rather than 16 or 24 bit? Well, you can create 32 bit float files from the outset if you want. You do this by setting the bit depth of your session to 32 bit float before you record. That way, you will capture your audio at the fixed point bit depth that your interface is capable of, but it will be saved as a 32 bit float file.
But that’s not really necessary. Here’s why… All modern DAW’s process audio internally at 32 bit floating point regardless of the bit depth of your audio files. As such, you’ll get the benefits of 32 bit floating point processing in your DAW, even if your audio files are not 32 bit floating point files themselves. As such, I recommend that you set your session to 24 bit when you record, rather than 32 bit float. You’ll still get the benefits of 32 bit floating point processing, but you won’t increase your file sizes.
Bouncing at 32 bit floating point
You also have the ability to bounce down audio as a 32 bit floating point file. As I mentioned earlier, this should only be used if the audio file is going to be imported into a DAW. It’s not suitable for the final bounce that you will upload to a digital distributor.
Bouncing 32 bit float files gives you an advantage that a fixed point bit depth file does not. With a 32 bit float file, if the file is clipping past 0dBFS, the file can be imported into a session and ‘unclipped’. That’s because the audio which has peaked past 0dBFS has not been lost. It is still there and can be played without distortion by attenuating the clip gain to a level where the loudest peaks do not pass 0dBFS.
With a fixed point bit depth however, this is not possible. If audio is bounced to a fixed point bit depth and there is clipping, then the clipped audio would be lost and any resultant distortion is irresolvable when imported into a DAW. Of course, if you’re gain staging properly and not being careless with your levels, then you shouldn’t really be in a situation where you bounce something and later realize that it clipped. But 32 bit float can offer that safety net.
An additional benefit is that when you bounce to a 32 bit floating point bit depth, dither is not required. So, if you’re going to be bouncing your audio a few times by moving it from session to session, then 32 bit float offers you the ability not to have to keep applying dither to the audio.
A word of warning
Something about 32 bit floating point that’s worth considering is this… Although you have the ability with 32 bit floating point processing to let your audio go past 0dBFS in your DAW without clipping, I still recommend that you treat 0dBFS as a ceiling. This is for two reasons.
Firstly, you may have older third party plugins that you use which don’t operate at 32 bit float. As such, these plugins will clip if the signal goes over 0dBFS.
Secondly, some plugins have a ‘sweet spot’ to consider. This generally occurs in plugins which are designed to model analogue equipment. As such, you will achieve different tonal characteristics based on the level at which you send signals into the plugin. Pushing really high levels into these kinds of plugins is not likely to result in the best sound, even if they operate at 32 bit float.
Can you record audio at 32 bit floating point?
Very recently, a couple of audio recorders have come to market which offer the ability to record at 32 bit floating point. One such recorder is the Zoom F6. This is a very new advancement in the field of audio recording and is currently available in portable field recorders. Will this ability expand into audio interfaces for recording studios? Time will tell.
These recorders combine multiple 32 bit fixed point convertors and then save the captured audio as a 32 bit floating point file. Because of the large amount of dynamic range, they offer the ability to record without worrying about your gain settings. The levels can then be scaled up or down as necessary in post-production.
The ability to capture audio without having to worry about gain could be very useful to people who record sounds with extremely high and/or unpredictable levels. Perhaps a nature recording which is unexpectedly interrupted by a sudden clap of thunder would benefit from this. Or perhaps the audio for an interview or podcast where there is no opportunity to ‘go for another take’ if you find that you didn’t set the gain correctly. For those of us recording musical instruments in recording studios however, would this be necessary? As long as you set your gain levels intentionally, and you’re recording 24 bit audio, then I can’t see this as being something that recording studio users should require.
That’s a lot of info to take on board. Let’s recap…
Bit depth dictates the dynamic range of digital audio by determining the level of the noise floor.
Record at 24 bit. This will give you more headroom to work with than 16 bit. There’s no reason to record at 32 bit fixed point as 24 bit is already more than enough. There’s also no real benefit to creating 32 bit floating point files when recording. That’s because your audio will be processed at 32 bit in your DAW, even if your audio files have a fixed point bit depth.
When it comes to bouncing, the only time you should bounce to 32 bit float is if the audio is going to be imported into a DAW. It’s not suitable for the final bounce. That needs to have a fixed point bit depth. The benefit of bouncing to 32 bit float is that if there’s any clipping, the audio can be attenuated when it is imported into a DAW and ‘unclipped’. 32 bit float also doesn’t require dither.
When you bounce your final track for release, choose 16 bit. 16 bit provides you with more than enough dynamic range to represent even really dynamic music. There’s no benefit to bouncing your final track at 24 bit.
What bit depth do you usually go for when you record and bounce your audio, and why? Leave your thoughts and suggestions in the comment box below.
Get the best results from EQ, compression, vocals & drums with the FREE 'Home Studio Bundle'
Hi Alex, I wanna ask you some question, if you create stem for mixing which bit depth do you prefered?
Hi Dan, thanks for your question. Personally, I find that most people mix at 24 bit, so I’d create it at 24 bit. I hope this helps!
Well explained and totally agree with your article. One minor point though, while an analogue signal is a continuous signal, it is incorrect to characterise it as having an infinite resolution. It that were the case it would also have an infinite dynamic range. In fact, it is the opposite way round, even 1 bit digital has potential of infinitiy if it wasn’t for the noise generated by quantitisation or dither. There is no analogue recorder or playback device that is capable of higher resolution than 16 bits.
Hi Kevin, thank you for your comment, and you’re quite right. My intention was to explain analogue signals as continuous. But I can see how the terminology I used to try and describe this could be taken to mean something different. I’ve edited the article to avoid any confusion. Many thanks!
It clarified my doubts… Thank U