1. Home
  2. misc-audio
  3. 195864-117078

Here’s how good AI is now. Check it out!

AI Overview

+18

Yes, the threshold for hearing audible phase shift on an analog equalizer can change significantly depending on the audio system and listening environment

. The audibility of the EQ's phase shift is influenced by factors beyond the EQ itself, including room acoustics, speaker design, playback volume, and the complexity of the audio material. 

How the audio system affects audibility

Room acoustics

  • Acoustic masking: In a typical listening room, sound reflections from walls, ceilings, and other surfaces create a complex web of constructive and destructive interference (a "phase soup"). This ambient phase distortion is far greater than what an analog EQ introduces, effectively masking the more subtle EQ-induced shifts.
  • Threshold shift: Because our ears are evolutionarily adapted to filter out the natural phase shifts caused by environmental reflections, the audible threshold for an EQ's phase shift is much higher in a normal room compared to a controlled, anechoic (echo-free) environment. 

Loudspeakers

  • Crossover effects: The crossovers in multi-driver speakers are a major source of phase shifts, as different frequencies are routed to different drivers (woofers, tweeters). The phase shifts caused by a speaker's own crossovers can easily dominate and obscure any shifts introduced by an external EQ. This is less of an issue with headphones or single-driver speakers.
  • Driver alignment: In multi-driver speakers, if the drivers are not perfectly time-aligned, it can create additional, audible phase distortion. Some high-end speakers, for instance, are designed to have all frequencies leave at the same time to ensure better phase coherence. 

Playback volume

  • Audibility increases with volume: Studies show that phase effects, particularly in the lower frequencies, become more noticeable at higher sound pressure levels (SPLs). If you are listening at low volumes, phase shifts are less likely to be audible, but they may become detectable when listening at higher, more intense levels. 

Headphones vs. speakers

  • Headphones offer higher sensitivity: Controlled studies have shown that phase distortion is more detectable on headphones than on loudspeakers in a normal room. Without the masking effect of a reflective room, the subtle changes introduced by an EQ's phase shift become more apparent. 

Complexity of the audio material

  • Musical content vs. test signals: Phase shift is generally inaudible on complex musical and speech signals. It is most easily detected with specific, simple test signals, such as square waves or tone bursts, and only under very controlled conditions. A phase shift that is obvious with a simple test tone may be completely lost in a dense mix. 

EQ-related factors that remain constant

While the audible threshold changes, the physical phase shift caused by a given analog EQ remains constant regardless of the audio system. The EQ's circuit design dictates the filter's minimum phase characteristics, which means the amplitude and phase response are inherently linked. What changes is the listener's ability to perceive that shift based on the overall context of the playback chain. 

tlcocks
thriftyaudio's avatar
thriftyaudio
192 posts
 

AI is just as smart as the material it is referencing. I have found to follow up on the data links in order to see how the conclusion was reached. As we all know, there is a bunch 'junk' on the internet, and AI does not filter it unless you purposefully remove the data from the response. I have received significantly different AI recommendations once poor (or good) sources are removed from consideration.

It is kind of like history text books. Those from the 50's are much different from the 70's, 90's, and even today.

tlcocks OP
568 posts
 

True. Good points. It DOES do well researching complex queries 

jrwaudio
365 posts
 

I've had AI for decades,  ever since I got married 

lanx0003
1,969 posts
 

+1 ... It DOES do well researching complex queries

Even though it makes stupid mistakes from time to time, ChatGPT outperforms Google Gemini and actually helped me resolve two issues I was having with the Volumio Rivo Plus that Volumio technical support (TS) could not figure out.

First, I used a USB cable to connect the Volumio Rivo+ with the Topping D90 III Discrete. Every possible setting was configured according to the manual and Volumio technical support’s instructions, and the D90 correctly displayed the bit and sample rates—indicating it was receiving the audio signal—but there was no sound. This back-and-forth went on for several days without resolution.

As a last resort, before completely giving up, I asked ChatGPT by typing in all of these symptoms. It did a long pause—almost like thinking or diagnosing for a full minute—and then produced a list of possible causes. Most of the suggestions were similar to what the Volumio support rep had given me, but one was new: the AI suggested that the 5V setting for the XLR port (which I was using) in the analog output stage might not be compatible with some sources or streamers. After I switched it back to the normal 4V setting, the sound came up.  How could the AI possibly know that?

Second, I noticed that at the beginning of playback for some songs—particularly classical music—the volume would suddenly spike and be accompanied by white noise. This had never happened before with any of my gear or system. I started to panic but calmly went through a step-by-step diagnosis to isolate the possible source of the issue: Qobuz or Volumio (software), streamer, DAC, or amplifier (hardware). After spending a lot of time and effort, I was fairly certain the cause was the Volumio app but couldn’t identify the exact culprit.

Then the AI suggested that the issue might be due to ‘Volume Normalization’ or ‘Resampling’ being turned ON. I disabled both settings, and the problem disappeared. Its troubleshooting capability is truly top-notch.

erik_squires
17,566 posts
 

Thanks for the kind words,  but even when I have found AI quoting me directly the "research" it does can be bogus or poorly sourced.  I’ve spent the last 2 months or so using it for research and coding and writing assistance.  You have to be really careful in leveraging it or you’ll be caught in one of many traps. 

You should treat large pieces like what the OP quoted as "multiple loosely joined opinions from the web" until you have narrowed down exactly where they came from.  AI does a poor job of separating "some bro said this" vs. actual research by experts. It’s really good for learning concepts, like ask it to explain group delay, then asking it questions about the math are usually accurate, or at least you can find sources in text books or articles from known experts.  

Ask it to tell you why silver cables sound better and you are going to get a lot of nonsense, which excludes contradictory opinions and evidence. 

Recently here I was pointed to a blog which appears to be an AI summary of a single research paper by an undergraduate.  The paper IS very interesting, but it’s 1 paper that has not been peer reviewed or put into the context of past work.  The blog author posted the summary as fact and failed to credit the source.  That these are immutable facts that should be accepted by all, and that’s where the real danger of AI is. 

When @audiokinesis writes something here, he is usually very clear about what prior work he’s basing his opinions on, and also makes it clear where and why his opinions/experiences have diverged.  That’s a model I hope to follow and I hope critical readers pay attention to when it’s missing from civil discourse and AI summations, and the kind of thinking that allows conspiracy theories to thrive. 

I leave you all with these cautionary tales. 

https://techcrunch.com/2025/09/18/openais-research-on-ai-models-deliberately-lying-is-wild/

 

https://www.cnn.com/2025/09/05/tech/ai-sparked-delusion-chatgpt

 

https://www.cnn.com/2023/08/29/tech/ai-chatbot-hallucinations

More to discover