AI Should Fix Concert Acoustics, Not Secretly Rewrite the Singer

The most useful AI at a concert may be the kind nobody notices. That is also the problem: when software changes what reaches our ears, audiences deserve to know whether it is correcting the room—or rewriting the performance.

During Céline Dion’s Paris residency, French audio company L-Acoustics used Source Intelligence, a machine-learning system designed to isolate a singer’s voice from instruments, crowd noise, echoes and speaker spill in real time. The technology was deployed alongside a large, carefully distributed speaker system at the Plenitude Arena, a venue built for rugby and known for difficult acoustics. BFM TV reports that the system can reduce non-vocal components by up to 40 decibels, according to L-Acoustics.

That is not an AI singer. It is something more practical—and, in our view, more interesting: an adaptive layer between the microphone and the audience. But the industry should not expect “it only makes things clearer” to settle the ethical question forever.

The best concert AI is a better pair of ears

Live sound is a hostile environment for clarity. A microphone aimed at a vocalist also hears drums, guitars, backing singers, audience noise and reflections from the venue. Those unwanted signals can make lyrics harder to understand and increase the risk of feedback. Source Intelligence analyzes the incoming signal, identifies the vocal component and suppresses some of what does not belong there before the sound reaches the mixing console.

This is exactly the kind of automation entertainment needs. It does not ask a model to compose a song, imitate an absent performer or decide what the audience should feel. It tackles an engineering problem that becomes harder as venues grow larger and productions become more elaborate.

The practical benefit is not just a cleaner vocal. L-Acoustics says Source Intelligence operates with latency below 8.5 milliseconds and can reduce non-vocal sound by as much as 40 decibels. Those are manufacturer-provided figures, not independent measurements taken during Dion’s Paris shows, as Clubic notes. That distinction matters, especially when “AI” is being used as a shorthand for a technical claim.

The broader lesson is that AI can improve access to the performance that is already happening. If a listener in the highest seats can understand the singer more easily, the system is correcting an uneven physical environment—not manufacturing a new artist.

The hardware still matters more than the magic label

It would be a mistake to describe the result as software saving a bad concert. Source Intelligence works within a complete sound design. For Dion’s shows, roughly 90 L1 speakers were distributed around the stage and above the floor area, creating more controlled coverage than a few giant speaker stacks could provide. The arrangement also helped keep the stage quieter, with reports describing a difference of up to 26 decibels between the stage and the audience area.

That quieter stage is an underrated part of the story. Performers can rely more on in-ear monitors rather than competing with sound bouncing back from the venue. Engineers get fewer unwanted signals in the microphones. The audience gets a more consistent mix from seat to seat. AI is one component in that chain, not a replacement for acoustic design, speaker placement or human judgment.

This is where the technology looks most convincing. The software handles repetitive separation and cleanup, while engineers retain responsibility for the artistic mix. L-Acoustics says the system is intended to support, rather than replace, those professionals. The company’s Denis Savage, Dion’s front-of-house engineer, described the effect as “like voodoo practically” in comments reported by The Straits Times. The colorful description is understandable, but the underlying achievement is less mysterious: better signal separation, delivered quickly enough for live performance.

A clearer voice is not the same as an untouched voice

Here is where the concert business needs sharper language. “Cleaning” a vocal can mean removing feedback and bleed. It can also mean changing timbre, dynamics or imperfections that are part of a particular performance. Those are not equivalent interventions.

The documentation describes Source Intelligence as isolating a real voice and attenuating unwanted sound. It does not describe the system as generating Dion’s voice or recreating notes she did not sing. That makes this use meaningfully different from synthetic performance technology. Still, audiences cannot independently hear the signal before processing, and the system’s work is intentionally invisible.

Transparency therefore should not be treated as an accusation. A concert ticket does not require a technical manual, but it should be possible for artists and promoters to state whether AI-based vocal separation, pitch correction, voice enhancement or generative replacement is being used. Those categories carry different creative and ethical implications. Lumping them together under “AI sound” is convenient for marketing and unhelpful for trust.

There is also a commercial angle. Source Intelligence works exclusively with L-Acoustics systems and requires the company’s L-ISA Processor II, tying the software to the manufacturer’s wider hardware ecosystem. The system used for Dion’s concerts cost more than €1 million as part of the complete audiovisual and acoustic installation, not as the price of the AI software alone. This is sophisticated infrastructure, not a plug-in every touring act can casually add.

The audience should get clarity about the clarity

L-Acoustics is extending a hardware business into software, and that strategy makes sense. The company says its technology is also being used by Harry Styles, Bruno Mars, The Weeknd and Rosalía. Other audio firms, including Alpha Labs, Cedar Audio and D&B Audiotechnik, are pursuing machine-learning tools for source separation, feedback control or virtual sound modeling.

That competition could make adaptive concert audio more capable and more common. It could also make the boundary between correction and alteration harder to see. Our preferred standard is simple: use AI aggressively to manage rooms, reduce noise and protect intelligibility; disclose it when it materially changes the singer’s sound or performance.

Céline Dion’s Paris shows offer a persuasive case for the first category. The technology appears to be helping a real performer reach a difficult venue more evenly, while leaving the artistic decisions to the human team. That is a better future for AI in entertainment than replacement—and it deserves to stay that way in the open, not merely behind the curtain.