Use of AI tools to improve one’s audio system

I recently queried three different AI agents—Microsoft’s Copilot, ChatGPT and Perplexity—regarding the efficacy of using LessLoss Firewall for Loudspeakers in my system. I had been using LessLoss filters for many years but having recently changed my amplification from a Kinki EX-M7 Class A/B to a Mola Mola Perca Class D, I thought I’d check whether the accessory was worthwhile.

Spoiler alert: Copilot seems leagues ahead of the other two AI agents, which I’ll explain below.

Here was my query to the AI agents: “I have a Mola Mola Perca amp connected by Grimm SQM XLR cables to a Lumin X1. The amp is connected to my Vivid B1 Decade speakers with Kinki Earth cables. Would the insertion of LessLoss Firewall for Speakers between the speakers and Kinki Earth cables help improve the sound quality?”

Both ChatGPT and Perplexity advised me to try the LessLoss filters. Both could be summarized as follows: “So: in a system like yours, the LessLoss Firewall for Speakers is likely to yield an audible improvement in smoothness, micro‑detail, and soundstage, but it’s not guaranteed or night‑and‑day; a home trial is strongly recommended before committing.”

On the other hand, Copilot seemed to better understand my system and advised against its usage. Here is a summary of its extensive comments:

“Your system already operates with exceptionally low noise and high transparency: the Lumin X1’s optical isolation (note: I had previously informed all the AI agents that the X1 was connected by fiber), Grimm SQM XLRs, and Mola Mola Perca’s ultra‑low‑impedance output stage leave almost no high‑frequency contamination for an inline device to ‘clean up.’ The LessLoss Firewall for Speakers works by adding series impedance to suppress HF noise, but in a chain as electrically quiet as yours, there’s simply nothing meaningful for it to address.

“The Vivid B1 Decade is also an unusually benign load—smooth impedance, minimal phase rotation, and very fast transient behavior. When you insert a device like the Firewall, its added resistance and inductance tend to soften leading edges rather than improve clarity. With speakers that thrive on speed and articulation, this usually translates into slightly slower transients, reduced micro dynamics, and a mild smoothing of upper harmonics.

“Because your Kinki Earth cables are already low‑resistance and low‑inductance, adding a filtering block in series is more likely to subtract than enhance. In this specific system, the Firewall isn’t solving a real problem and may gently dull the qualities your setup excels at: immediacy, precision, and micro dynamic nuance.”

Obviously, my test was a small sample, though I did use Copilot to assess the efficacy of my system within the Roon ecosystem and ended up enabling volume control to complement Leedh Processing in my Lumin X1.

Last point: your mileage may vary and the inconsistent results of various AI agents may make it difficult to rely on any one, but I think it’s worth experimenting. In the end, however, your ears will prove to be the best judge.

1 Like

All these “AI” agents do is summarize information that a human wrote on the internet, correct or not, and then the “AI” might well invent other things on top of it out of thin air.

4 Likes

When possible, AI results need to be fact checked.

Two weeks ago coworkers and I discussed a certain old DAC chip we used in the past - WM8741. A coworker queried his favorite AI engine for a particular sample rate specification of it (instead of searching for the data sheet and reading it), and it differs from my memory. I am right, and the AI was wrong. I was actually surprised how AI could give a wrong result from such a simple direct look up from data sheet.

Another coworker queried another AI engine for a tourist visit to different vinyl shops in a certain country. The AI hallucinated half of those shops, which are non-existent.

Also a noteworthy news with an important lesson:

1 Like