Search for DSP audio developper

[Real Mono Sound]

Hi,
We are looking for a DSP audio developper to make a Windows app.

Our project :

The job is to code a +90° phase shift (similar to Voxengo PHA-979 set to +90°) on the L - R signal before adding it to the L + R signal.
A Waves L2 type limiter is also needed to limit the new transients.

Contact : d.holly@realmonosound.com

[Real Mono Sound]

Full description https://www.facebook.com/share/p/19KZ7xTs1R/

[Real Mono Sound]

Teaser https://www.facebook.com/share/v/1DRdQ6FY4F/

[Real Mono Sound]

Real Mono Sound speaker : the ultimate solution for mobility https://www.facebook.com/share/p/1NEWTAo9iE/

[Real Mono Sound]

We are looking for partnerships :

- to implement our process in apps and operating systems,

- to make hardware equipment for professionals,

- to make speakers containing a DSP chip featuring Real Mono Sound processing.

Our European patent will be extended to the US in the near future.

Contact : d.holly@realmonosound.com

[mystran]

Watch out that a 90 degrees phase shift can cause essentially unbounded increase in peak levels. Worst-case signals are things like bright saw or square waves that theoretically cause peaks to become infinite (in practice they will be finite due to digital audio being bandlimited, but they are still get huge). This in turn means you'll potentially be hitting the limiter very hard, causing distortion as a result.

Such phase shift also tends to make a mess out of transients. You can hear this clearly in the "rock" example in the video from the first post, where traditional mono is nice and tight (although the guitars probably have some effect that isn't really mono compatible so they lose some power) where as the "real mono" sounds kinda washy.

I'm sure some people are going to like it simply because it sounds "different" but .. frankly I'd rather take a good traditional mono mix over this every time.

[Real Mono Sound]

Thanks for your message.

Yes, there is sometimes increase in levels but it's mostly in the +3dB range. A L2 type limiter can easily handle this.

I can't hear any audible artefacts in phase, frequency or transient response. I'm using the Voxengo phase shifter. It's easy to replicate in a daw : duplicate a stereo file to two groups, make the first group mono, make the second group stereo only - using Voxengo MSED for example - and apply the phase shift - using the PHA-979 plugin for example - on the stereo content.
Anyone who tries it will find it rather transparent and natural I believe.

The rock example you mentionned sounds a bit washy in the first place, maybe that's why. Again, there is no coloration to the processed signal compared to the original stereo signal 99% of the time.

I agree that a mono *mix* is fine if it means the track has been mixed in mono. It's better than summing a stereo mix to mono.

Another point I'd like to mention is that sound engineers and producers can now use the stereo content creatively, instead of ensuring that it's disappearance in mono does not critically affect the audio message. For example, they can increase the stereo content/effect while still achieving faithfull mono playback.
In other words our innovation makes any stereo mix fully mono-compatible.

[Real Mono Sound]

Here is some chatgpt information :

To achieve an approximately constant 90° phase shift across a wide frequency range (20 Hz–20 kHz), it is generally necessary to use:
a cascade of several carefully optimized all-pass filter sections; or
a Hilbert-transform FIR filter.

[signalsmith]

Our European patent will be extended to the US in the near future.

Did you actually get granted a patent for this idea, or just apply for one? I'm hardly one to talk, but getting a patent for a (fairly small) algorithm like this shouldn't work.

BTW, this idea works if the stereo elements are independent or mono-panned (or somewhere in between). It falls apart a little bit if there's already a constant non-zero phase shift between L/R (or equivalently M/S).

My classic problematic example for this would be close-stereo-mic'd piano, because the range of string positions mean there's always a range of phase relationships, and the best you can aim for is that none of the fundamentals are fully phase-inverted. Do you have something similarly awkward in your test set?

[mystran]

Here is some chatgpt information :

To achieve an approximately constant 90° phase shift across a wide frequency range (20 Hz–20 kHz), it is generally necessary to use:
a cascade of several carefully optimized all-pass filter sections; or
a Hilbert-transform FIR filter.

Assuming you want to keep the original mono "as is" then you really want to use a Hilbert FIR. Unfortunately if you want this to work at low frequencies the filter will have to be fairly long. If you decide to go this route, one option would be to combine (convolve together) the Hilbert FIR with a high-pass somewhere in the lower mids, which would help avoid the processing making a mess of the most sensitive bass region (ie. just process everything above; bass is fairly commonly mixed more or less mono anyway) and also help lessen the pressure on the Hilbert FIR length. Unfortunately linear-phase FIR high-pass filters aren't exactly low latency either.

All-pass filters have the advantage that you can use a warped elliptic design to get a very steep transition at the low frequencies (at the cost of shallower transition at the less critical high frequencies) with fairly low orders, but they have the downside that even though the output will be quadrature (ie. a pair of signals at 90 degrees phase relative to each other), neither signal will actually be in phase with the input (ie. there is some overall phase-shift), so you'd have to apply the filtering to both signals (so it won't be transparent even for pure mono input then).

Usually I'd recommend quadrature all-pass filters for applications (eg. frequency shifters and such) that are tolerant of some phase smearing due to how they don't need a ton of latency to get a good frequency response.

ps. I'd estimate the practical worst-case peak amplitude increase (assuming full-band processing) to be on the order of perhaps 12dB or so. This would not occur in all styles of music, but things like "rude" EDM bass-lines and such could potentially cause such peaks when phase-shifted.

[alexeykoch]

My AI helper showed that this shift can be done in two formulas, which is fairly simple using FFT.

[Zeisner]

Not every stereo mix is ILD-only. Especially in big band/orchester music (but in some jazz/folk/country recordings too) you're dealing with a mixture of ILD and ITD (=equivalence stereophony). You can't just twist the phase of such a side signal and mix it with the mid signal without creating audible comb filtering. Your idea is also incompatible with ILD mixes with special M/S EQ settings (see Blauert's ribbons). And it isn't new either - I read about it more than twenty years ago in an engineering magazine. There's also Ambisonics which would require much less processing power than your idea (=cheaper) but isn't used as the standard for everything either because the additional costs still outweigh the benefit.

So here's an advice before you put any real money into this project: Drop it. It's better to be disappointed now than tens of thousands of Euro later. No major hardware or software company will consider your magic mono button if it won't work on everything. All it takes is a test with one incompatible track.