Nemisindo, our new spin-out, launches online sound design service

We haven’t done a lot of blogging recently, but for a good reason; there’s an inverse relationship between how often we post blog entries and how busy we are trying to do something interesting. Now we’ve done it, we can talk about it, and today, we can launch it!

Procedural audio is a big area of research for us, which we have discussed in previous blog entries about aeroacoustics, whistles, swinging swords , propellers and thunder. This is sound synthesis, but with some additional requirements. Its usually intended for use in interactive content (games), so it needs to generate sound in real-time, and adapt to changing inputs. 

There are some existing efforts to offer procedural audio. However, they usually focus on a few specific sounds, which means sound designers still need sound effect libraries for most sound effects. And some efforts still involve manipulating sound samples. Which means they aren’t truly procedural. But if you can create any sound effect, then you can do away with the sample libraries (almost) entirely, and procedurally generate entire auditory worlds.

And we’ve created a company that aims to do just that. Nemisindo, named after the Zulu for “sounds/noise” offer sound design services based on their innovative procedural audio technology. They are launching a new online service, https://nemisindo.com, that allows users to create sound effects for games, film and VR without the need for vast libraries of sounds.

The following video gives a taste of the technology and the range of services they offer.

Nemisindo’s new platform provides a browser-based service with tools to create sounds from over 70 classes (engines, footsteps, explosions…) and over 700 preselected settings (diesel generator engine, motorbike, Jetsons jet…). It can be used to create almost any sound effect from scratch, and in real-time, based on intuitive controls guided by the user.

If someone wants a ‘whoosh’ sound for their game, or footsteps, gunshots, a raging fire or a gentle summer shower, they just tell the system what they’re looking for and adjust the sound while it’s being created. And unlike other technologies that simply use pre-recorded sounds, Nemisindo’s platform generates sounds that have never been recorded, a dragon roaring, for instance, light sabres swinging and space cannons firing. These sound effects can also be shaped and crafted at the point of creation by the user, breaking through limitations of sampled sounds.

Nemisindo has already caught the attention of Epic Games, with the spinout receiving an Epic MegaGrant to develop procedural audio for the Unreal game engine. 

The new service from Nemisindo launches today (18 August 2021) and can be accessed at nemisindo.com. For the first month, Nemisindo is offering a free trial period allowing registered users to download sounds for free. After the trial period ends, the system is still free to use, but sounds can be downloaded at a low individual cost or with a paid monthly subscription.

We encourage you to register and check it out.

The Nemisindo team can be reached at info@nemisindo.com .

The crack of thunder

Lightning, copyright James Insogna, 2011

The gaming, film and virtual reality industries rely heavily on recorded samples for sound design. This has inherent limitations since the sound is fixed from the point of recording, leading to drawbacks such as repetition, storage, and lack of perceptually relevant controls.

Procedural audio offers a more flexible approach by allowing the parameters of a sound to be altered and sound to be generated from first principles. A natural choice for procedural audio is environmental sounds. They occur widely in creative industries content, and are notoriously difficult to capture. On-location sounds often cannot be used due to recording issues and unwanted background sounds, yet recordings from sample libraries are rarely a good match to an environmental scene.

Thunder in particular, is highly relevant. It provides a sense of the environment and location, but can also be used to supplement the narrative and heighten the tension or foreboding in a scene. There exist a fair number of methods to simulate thunder. But no one’s ever actually sat down and evaluated these models. That’s what we did in,

J. D. Reiss, H. E. Tez, R. Selfridge, ‘A comparative perceptual evaluation of thunder synthesis techniques’, to appear at the 150th Audio Engineering Convention, 2021.

We looked at all the thunder synthesis models we could find, and in the end were able to compare five models and a recording of real thunder in a listening test. And here’s the key result,

This was surprising. None of the methods sound very close to the real thing. It didn’t matter whether it was a physical model, didn’t matter which type of physical modelling approach was used, or whether an entirely signal-based approach was applied. And yet there’s plenty of other sounds where procedural audio can sound indistinguishable from the real thing, see our previous blog post on applause foot .

We also played around with the code. Its clear that the methods could be improved. For instance, they all produced mono sounds (so we used a mono recording for comparison too), the physical models could be much, much faster, and most of the models used very simplistic approximation of lightning. So there’s a really nice PhD topic for someone to work on one day.

Besides showing the limitations of the current models, it also showed the need for better evaluation in sound synthesis research, and the benefits of making code and data available for others. On that note, we put the paper and all the relevant code, data, sound samples etc online at

And you can try out a couple of models at