Deconvolution Dereverberation
As a music producer, I am well aware of what seems to be an as-yet unsolved problem, but one which is soluble (at least partially).
When I use samples, especially orchestral samples, they are usually presented with lush, ambient reverberation to make them sound gorgeous, so that you’ll buy the sample library. Orchestral samples are frequently recorded in some famous concert hall or other, to add prestige to the product and these halls are generally very lively, to enhance the sound of the orchestra to a live audience in the auditorium. Without a full house, however, which is what happens during sample recording sessions, there are no human bodies in the audience to soak up the room reflections, so the reverberation captured on the samples is even more lively than during a typical live concert.
Indeed, on their own these recordings sound gorgeous, dripping wet with characterful room reflections and diffusion, but the problem arises when you place those samples in a mix. Invariably, the reverberant wash is overwhelming and you want to dry the sound out, taking some of the reverberation away, but you can’t. The reverberation is baked into the sample recording.
Consequently, your choices are poor. Either you can smear the orchestral sounds with equalisation and gating, to eliminate some of the reflections, or you can attempt to use one of the many de-reverberation plug ins to automate this process, or else you can end up with an overly reverberant mix that sounds unbalanced, or do without the samples altogether. None of these choices are very satisfying, creatively. What you ideally want to do is to remove the reverberation and perhaps add a different reverberant character to your recordings, so that they sound like they’re in a more cohesive sound field, along with your other instrument parts.
Every room (e.g. a concert hall) has an impulse response. You can hear it by clapping your hands once and listening. That series of flutters and echoes, which decay into a diffuse sound field, is the impulse response of the hall, as near as dammit. Those skilled in the art of digital signal processing know that if you take a dry instrument recording and “convolve” it with the recorded impulse response of the concert hall, you can make the instrument sound like it was played in that hall. Convolution is basically multiplying one signal with another. Change the impulse response and you change the hall. The essence is that you need to record the instruments with very little room colouration, but sample library vendors resist this, because instruments recorded drily sound awful. Everything needs a little reverberation to sound natural. If the sample library is presented dry, it’s unlikely to attract buyers, the vendor can’t control the acoustic you hear the sounds in and the sampled instrument players will be put off by the horrible acoustic they are recording in.
There is another technique well-known to digital signal processing engineers called “deconvolution”. The theory goes that if you know the impulse response of the room, you can dry the recording out by taking the room sound away from the recording mathematically, leaving a simulation of an instrument recorded drily.
So, a way forward is to present reverberant samples the way are recorded today, but ensure that an impulse response of the room is also recorded, allowing that to be deconvolved from the composite recording. What you do is make a single, loud spark, where the instrumentalist is usually standing or sitting and record that with the same microphone array you’re using. If sample libraries and orchestral recordings included this impulse response recording, then with the magic of digital signal processing (e.g. in a Digital Audio Workstation plug-in), the impulse could be deconvolved from the instrument recording and a dry signal would become available to the music producer.
There is a wide range of choice in artificial reverberation solutions, including algorithmic ones and convolution reverbs. Once the signal has been dried out, it can be re-wet with greater control over reverberant characteristics. This also works with music recorded in poor acoustic environments, not just sample libraries. The linear response of the lousy room the musical performance was captured in can be deconvolved and a better one superimposed, via convolution of other algorithmic means. This could rescue many wonderful performances recorded in small clubs, badly designed studios and in bedroom home studios.
These deconvolution techniques have their limitations. They’re not perfect. The impulse response won’t ever be exact. Not all acoustic colourations are linear. There will be some reverberation artefacts that you simply cannot remove, no matter what you do. However, it’s a far better option than the current attempts at de-reverberation, which are not based on deconvolution of the actual room impulse response, but on heuristic post-production techniques.
This is a product that, if available, I would definitely use. It would require the co-operation of vendors of royalty free samples to play ball and provide the impulse responses of the halls they record in. That could present it’s own difficulties, of course. Abbey Road studios, for one, prohibits the recording of impulse responses in their hallowed and famed recording studios. To allow this would be to dilute their unique selling point as a studio. The sound of their rooms is distinctive, desirable and valuable to them, as a business.
That said, the technique is viable, at least in theory and worth pursuing. Many live venue sound enhancement devices work by deconvolving microphone signals from the room impulse response in real time, to make it possible for live sound engineers to “dry out” their live presentations on the fly. Taking out some of the reverberation captured by the microphones out, in effect, prevents the signal from being reverberated twice — once when captured by the live microphones and again, when reproduced by the public address loudspeakers. An audience only needs one pass of reverberation.
I’d like to hear from some other music producers as to their opinion about this possible product. Would you use it?
About the author
Michael Topic is a freelance Product Manager with over thirty years experience delivering products that didn’t exist before. He welcomes contract enquiries to define new, competitive products, design them and deliver them. His speciality is software-based products.
Disclaimer
The hypothetical design ideas discussed are intended to demonstrate possibilities in product design. They are not intended to imply anything whatsoever about past or present employers, nor to infringe on any intellectual property.
About the “Possible Future Designs” Series
This occasional series of articles examines the many ways in which product management discipline can be applied to a variety of markets, to propose and examine innovative, new product opportunities.
Organisations wishing to pursue any of the ideas discussed are encouraged to contact the author to discuss potential research and development collaboration.
