Our NVH toolkit was built to take sound away. Now that we put sound in on purpose, we need to get much better at testing something none of our instruments can actually see
For most of my career, my job had a very simple definition of success: less. Less noise, less vibration, less harshness. And we got extremely good at it. Sound pressure level, loudness, sharpness, roughness, transfer path analysis – mature, standardized, reliable tools, all built to answer one question: how much unwanted sound is there, and where is it coming from?
Then the engine went away, and so did the premise. In a high-performance EV, a lot of what the customer hears is there because we decided to put it there. Sound stopped being a defect and became a feature. The moment that happened, we inherited a problem I don’t think our profession has really faced up to yet.
Here’s the simplest way I can put it: excitement has no unit of measurement.
I can measure an EV’s active sound at 60km/h to 1⁄₁₀ of a decibel. I can give you its loudness, its sharpness, its tonality, its order content. What I cannot give you is whether the driver felt anything. When we were developing the driving sound for the Ioniq 5 N, there was no target to converge on, because none exists. We weren’t minimizing a quantity, we were trying to provoke a reaction. Normal NVH work hands you a number and a direction. This hands you neither.
That is hard enough. But there’s a second issue underneath, and I think it’s the more serious one.
Our instruments measure channels separately. A microphone takes the sound. An accelerometer takes the vibration. Vehicle dynamics and powertrain control sit in different data streams, usually looked after by different teams, quite often in different buildings. Each domain gets validated against its own targets and signed off on its own terms. But the person in the driver’s seat doesn’t experience channels. They experience one event.
This isn’t a philosophical point. When we were integrating sound with virtual gearshift and haptic feedback, what separated a convincing shift from an unconvincing one usually wasn’t the quality of any single element – each was excellent on its own. It was the alignment between them. Let the audible event and the physical one drift apart by a few tens of milliseconds and the whole illusion falls over. The driver can’t tell you what’s wrong. They just say it feels fake.
Now think about how that gets caught in a normal test campaign. The sound passes. The haptics pass. The control strategy passes. Every domain reports green, and the product is still wrong. We validated the parts and missed the thing we were actually selling.
So if I could change two things, they would be these. First, treat synchronization as something you measure. As far as I know, the timing relationship between sound, haptics and vehicle behavior is almost never written down as a requirement with a tolerance, instrumented, and signed off like any other engineering target. If the multisensory experience is the product, that relationship is a specification, not an artistic detail, and it belongs in the test plan with a number attached to it.
Second, start recording what people say. There is a lot of enthusiasm at the moment for AI that can predict perceived quality, and I share it – analysis of driver reactions, biometrics and expression will eventually give us a much better read on emotional response than we have today. But an algorithm can only learn a correlation someone bothered to record. In most organizations the objective data is archived beautifully and the human verdict that went with it isn’t archived at all. That is the real bottleneck. It isn’t the state of machine learning, it’s that we aren’t collecting the data those systems will need.
None of this needs technology that doesn’t already exist. It mostly needs a decision that the subjective half of the job deserves the same discipline we have always given the objective half.
We spent several decades getting very good at measuring what we wanted to remove. I suspect the next decade will judge us on how well we learn to measure what we chose to create.
