With so many new engineers turning solely to the internet for answers, ChatGPT’s AI has great potential for those learning about audio. Is it really a fountain of knowledge or just another way to get misinformation? We test it.

In Summary

ChatGPT’s Large Language model can only draw from information that has been written by real people, or other AI responses that can generate third-hand ‘ghost references’. In our unscientific test, we ask it a mix of biased and unbiased questions to see if it gives accurate answers. We then compare those to information from verified sources written by respected industry professionals.

Going Deeper

For anyone unfamiliar with it, ChatGPT is an online tool for analyzing and generating text. Using tech inspired by the human brain, they can generate apparently sentient responses (something they are not doing) using information gleaned from the internet and other AIs. With uses rooted in solving computer coding problems, it also has vast potential for creatives working in any media.

Here we present it with some common audio myths using the kind of questions that new engineers might ask. As such, we ask it some biased questions, as well as some that steer it away from its tendency to agree with everything! We’ve included the most significant sections of its answers, and we’ve quoted directly from these. The more general sections of its responses have been left out.

How Is Phantom Power Bad For Ribbon Microphones?

ChatGPT says:

Phantom power, which typically supplies +48 volts DC, can potentially be damaging to ribbon microphones, particularly older or vintage models and some passive ribbon microphones. The main concern lies in the design and construction of ribbon microphones, specifically their delicate ribbon element.

A professional opinion:

In our article, Derek Bargaehr states:

In theory, hot-swapping a ribbon with good XLR cables/connectors should not cause this issue as pin 1 (ground) connects 1st on most XLR cables. The exception is a damaged cable where there's a short between pin 2 or 3 and ground or a worn-out connector that isn't making great contact. With a good phantom power supply, on the other hand, there will be a slow ramp-up to the 48V, as opposed to a sudden 48V surge (i.e., the transient). This is all to say - hot-swapping can be fine, but is not ideal.

Verdict:

Phantom power is not always bad for ribbon mics, including some vintage designs. ChatGPT’s answer is based wholly in the ‘bad’ camp, and does not address the exceptions to the rule, including perhaps the most obvious case: active ribbon mics. The reality is far more nuanced, as covered by our panel of experts in our article on the subject.

Which Is Best For Recording Studios? Triple Glazing Or Double Glazing?

ChatGPT says:

Triple glazing typically provides better sound insulation compared to double glazing due to the additional layer of glass and air gaps. This can help create a quieter and more isolated recording environment.

A professional opinion:

In this article, Sam Inglis explains how, for a given total depth, triple glazing generally does not improve acoustic isolation:

The degree of isolation afforded by a double-glazed window, or indeed any ‘two-leaf’ structure such as a cavity wall, depends partly on the mass of the panels, and partly on the width of the air gap. In particular, as the air gap is narrowed, the effectiveness of the isolation at low frequencies is lost. What this means is that in a window of any given depth, you’ll get better low-frequency isolation by having two panes of glass separated by a wide air gap than by inserting an additional pane in the middle and creating two narrow air gaps. It is true that triple glazing can provide more isolation in the high frequencies, but these are almost never the problem!

Verdict:

ChatGPTs claim that triple glazing’s additional layer of glass and air gaps provide better sound insulation compared to double glazing is misleading. On the understanding that a window opening’s depth is fixed, double glazing’s effectiveness will be reduced by adding a third pane in the middle of two outer panes. Doubling the depth of the entire system to add an extra air cavity using three panes would however improve isolation.

Why Do I Need To Make My Music Master As Loud As Possible?

ChatGPT says:

The perception that music masters need to be as loud as possible is a common misconception in the music industry, often driven by the "loudness wars" phenomenon. While loudness can be desirable in certain contexts, such as commercial releases or genres that prioritize a high-energy, impactful sound, it is not universally necessary or beneficial for all types of music.

Many streaming platforms and broadcasting systems have loudness normalization algorithms in place to create a consistent playback experience. These algorithms adjust the volume of tracks automatically, often targeting a specific loudness level. Consequently, excessive loudness maximization during mastering may not have the desired impact when played back through these platforms.

A professional opinion:

In our podcast, audio mastering grandee Bob Katz has this to say of blanket loudness normalisation:

It's not the loudness as much as the quality that counts. If we look at a characteristic that we call Peak to Loudness Ratio (PLR). Although it is not an infallible determination of quality, it is a pretty good indication that the transients are getting through and if you squash the transients too much, then the quality gets disturbed.

Verdict:

ChatGPT correctly challenges ‘louder is always better’. While its assertion that over-loud tracks will lose impact on some services is true, it lacks the professional’s take on other markers of quality. By concentrating on Integrated Loudness and ignoring Peak to Loudness Ratio, ChatGPT’s answer runs the risk of ignoring what really counts: How It Sounds.

Should I Record As Hot As Possible? Why?

ChatGPT says:

In most cases, it is recommended to aim for a balanced approach during recording, where the signal level is healthy but not excessively hot. This allows for optimal signal-to-noise ratio, preservation of dynamic range, and flexibility in subsequent processing. Monitoring the levels carefully and ensuring that the audio signal stays well within the acceptable range, avoiding clipping, is key to capturing high-quality recordings.

A professional opinion:

In this article, Hugh Robjohns goes further by putting numbers into the technical argument for leaving generous headroom while recording:

Some will argue that allowing 20dB of headroom compromises the noise floor by 20dB. This is technically true, but completely irrelevant. The noise floor of even budget 24-bit digital converters will be better than -115dBFS, so leaving a very generous 20dB headroom margin still puts the noise floor -95dB lower than the new operating level. Comparing this to the analogue world, few high-end analogue consoles will have a noise floor below -90dBu, and with clipping at +24dBu the total dynamic range is an almost identical 114dB. In practice, then, the noise performance of a digital system — even with a 20dB headroom margin — is no worse, and will probably be slightly better than that of a high-end analogue system.

Verdict:

ChatGPT correctly asserts that recording hot is no longer necessary, however its reasons are vague. Hugh Robjohns’ assessment of the situation is far more useful compared to ChatGPT which struggles to say what “healthy but not excessively hot” actually looks like.

Final Thoughts

The results confirm that ChatGPT, for now, is able to turn in some inaccuracies, even when asked unbiased questions. With biased questions, its suggestibility is important, because it is less able to help newer engineers to get away from any misconceptions they might have. Asking it an unbiased question can however, sometimes return accurate results.

As the pool of knowledge grows, so too will the depth of ChatGPT’s responses. One fact remains, and that is that its current technology means that it can never actually know what it is saying. With the not-so-new problem of recycled information, and that of ghost references that lean on AI itself, perhaps the question of who is contributing to ChatGPT’s sources is the most significant one.

What do you think about the accuracy of ChatGPT? Let us know in the comments.

A Word About This Article

As the Experts team considered how we could better help the community we thought that some of you are time poor and don’t have the time to read a long article or a watch a long video. In 2023 we are going to be trying out articles that have the fast takeaway right at the start and then an opportunity to go deeper if you wish. Let us know if you like this idea in the comments.

Console photo by Maxwell Hunt on Unsplash

Meter bridge photo by Kenny Eliason on Unsplash

Window photo by Kenny Eliason on Unsplash

Pro Tools photo by Thibault Maestracci on Unsplash