Vehicle Engine design – Project CARS, Forza Motorsport 5 and REV
With this article I really wanted to find out about the nuts and bots of vehicle engine sound design and implementation. So I contacted a few people and got some great responses and a fascinating insight into the process. My thanks to Stephen Baysted, Audio Director and Composer at Slightly Mad Studios, Greg Hill, Sound Designer at Soundwave Concepts, Adam Boyd, Sound Designer and John Twigg, Software Engineer at Crankcase Audio and Nick Wiswell, Audio Creative Director at Turn 10 Studios.
DS: I started by asking Greg Hill about his work on racing titles in the 90′s.
Greg Hill: In the mid to late 90’s I was just modding the sounds in titles like Need For Speed, GP1, Sports Car GT & GP Legends. I’ve always been a musician with recording gear; I taught multimedia, I’m bit of a computer geek and a car fanatic – well – this all came together! Way back then CPU’s were getting faster and GPUs had 3DFX but RAM was an issue and super expensive for regular home systems. Typically, sound got the smallest slice of the resource pie, but there was a growing expectation by most that sound should be considered with much higher regard.
The early engine sounds were either recorded ramps of acceleration, tone generated (synthesized) or a couple of loops. The ramp system used recorded sweeps of acceleration (low RPM to high RPM). These where really just one-shots with simple triggers that just re-spawned with each gear shift, they had no direct correlation to RPM data). The tone generated engine sounds were pulses of grains that sounded like weed-whackers and the loops were usually pitch bent so far they sounded muddy at low RPM and chipmunk/pitchy at high RPM. These systems were limited and ended up sounded pretty bad! However, quality recordings implemented into these games did improve things a lot. So I sent my sounds to these studios and they sat up and listened to my crazy ideas about sonic modelling engines and other modulated vehicular sound events like transmission whine and tyre/surface sounds.
I pushed these developers to model engines in a more realistic way than ever done before. My first clients were Image Space Inc., Papyrus Design Group Inc. and Torus Games. Together we created the first multi layered sonic models for engine RPM and load that were parameterised by the parameterised physical events including the mechanical systems. I worked on Sports Car GT 2 and Grand Prix Legends 2 – both these first titles I worked on were scrapped for F1 2000 and the Sierra NASCAR Racing titles among others. But now the tech we created was well on its way and I even got to work with middleware developers on similar tech such as Analog Devices SoundMAX SMartTools SDK, and Firelight Technologies on FMOD.
DS: And now?
Greg Hill: We have more CPU power, RAM and DSP!
Today racing audio is highly interactive and not sitting passive as stems baked onto video. The gamer makes choices like the surface they drive on, which gear, the thresholds of grip, throttle inputs (throttle on, off, in-between), engine loads, RPM, speed, even the weather conditions and what view they drive in and review replays in. I’m not talking about interactive like pulling a trigger where you fire off one-shots of sound, I’m talking about something on a much higher level than just that. These cars make sounds that modulate, move, and transition from one threshold to another and everything in between – it’s the gamer’s choice and the fundamental sounds have to be there for every major event. With modulated sound there is no room for noticeable gaps as the timbre of the source changes and these we call sonic models; I liken it to VST Instruments. I now make vehicles sounds using real recordings that are broken down into smaller parts so they can be reassembled at run-time. These sounds also must be efficient and expected to playback fast, blip a throttle from idle to maximum RPM and many samples must seamlessly integrate for both the on throttle sounds when accelerating and the off throttle when you lift the throttle and decelerate.
DS: When I first got in touch with Greg about this article he mentioned the following;
FYI, there are basically two popular ways to create sonic models of engine sounds. Loops and Granular.
DS: So I asked him to elaborate.
Greg Hill: Today the best results are the culmination of 3 key skill sets:
Each one of these is a bit of a black art when it comes to racing game audio and this is done really well by a few studios. For many studios the recording and editing can be similar but the implementation can be vastly different. Today there are 2 main methods for implementation and these are crossfaded loops and granular synthesis (spectral analysis of the audio data and the additive interpolated synthesis of grains). Crossfaded loops remain the most widely used to date. The best results I believe are a combination of both these techs, this form of hybrid loops is something I do more at the editing stage rather than implementation. Implementation for most of my clients is still predominantly loop based. My loops are not strictly created from simple flat sections at a static RPM as some people would have everyone believe – this is where the spectral analysis of the audio data is very handy to me. With a crossfaded system I can take the best parts from a range of source and provide very realistic results in multi-channel formats (stereo, binaural, surround) and include audio data like resonances, exhaust baffle vibrations, crackles and pops that would otherwise confuse and scramble the tracking within a granular synthesis method. Hybrid loops have allowed me to further develop my new technique for external ‘distant’ engine sounds and this is a real breakthrough because it allows very complex sonic models that contain a lot of residual elements so they sound very realistic in replays and for AI vehicles. There are a few videos of this tech getting around on YouTube for iRacing, Project CARS, rFactor 2, The Next Car Game etc.
These crossfaded sonic models take longer to do, but it’s an efficient and scalable method for most gaming platforms and it remains the best solution for simulators as used by F1, NASCAR and racing teams. Granular synthesis is getting better but sadly the interpolation is pretty synthetic sounding in certain situations so the tech uses filters to mask these anomalies. If granular was the best I’d be using it exclusively. Don’t get me wrong I love the granular tech and it can sound awesome – just listen to the engines in the current Codemasters games. I have worked with teams on this tech but right now it is too limiting for serious simulations. I do feel that one day granular will be the best and then I’ll be all over it.
DS: Stephen Baysted takes up the point.
Stephen Baysted: On our current titles – Project CARS and World of Speed – we’re using FMOD. For each car there are well over one hundred separate wave files that together constitute a ‘sound set’. These are layered and crossfaded and ‘played’ by physics inputs. With Greg Hill (my long term colleague) we try to record every vehicle that is modelled in the game and for each recording we ask the racing drivers to undertake a specific set of tasks. These tasks give us the best chance of recreating the sonic characteristics of the real racing car. Ideal captures are usually obtained from controlled recordings on a chassis dynomometer, but in the majority of cases we record cars on the track – and this has numerous challenges – not least when they crash or equipment catches fire!
DS: Nick Wiswell takes up the point in Forza Motorsport 5
Nick Wiswell: FMOD Studio was a huge part of the success of the audio in Forza Motorsport 5 for several reasons.
• We prototyped many of the key game systems offline in Nuendo (our DAW of choice), then once we had the sound we wanted it was really easy to bring that system into Fmod Studio and have it up and running very quickly using the same principals and techniques we had used in Nuendo.
•The ability to able to build very complex signal flow pipelines in a very intuitive manor.
•The flexibility of the send/return system and the VCA controls gave us a huge amount of creative freedom to try new ideas and techniques we had not been able to do before on previous Forza titles.
Without this functionality and flexibility, Forza Motorsport 5 would have been a very different sounding game.
DS: So what about granular then? Cue Adam and John and their latest product REV.
Adam Boyd & John Twigg: The games industry has always been behind in terms of the availability of great tools for audio designers, and after many years in the industry, we felt there was a particularly urgent need for a better vehicle engine modeling system. We had two main goals; one was to avoid the artificial sound resulting from the standard loop-based approach; the other was to greatly reduce the complexity and time required to create a high quality engine model. We did a tremendous amount of research into spectral analysis and granular synthesis, and after about two years we had a tool and runtime we felt was ready to take to market. We also had some great content partners, like Pole Position Production, The Audio Guys and Bryan Watkins, who were willing to share and entrust us with their amazing recordings while we developed REV.
DS: Take us ‘under the hood’ of REV.
Adam Boyd & John Twigg: REV works by tracking the harmonic components of an engine “ramp”, i.e. an acceleration or deceleration recording. The sound is broken down into discrete slices at the granularity of individual pistons firing. Once we have separated the individual piston cycles, the ramp can be reproduced as it was originally recorded by concatenating the cycles back together – but more importantly it can be made faster by skipping over cycles, or slower by duplicating and repeating cycles. It can also be made completely static, or even be made to play backwards. The real power of REV is the ability to take an essentially fixed sound recording and make it pliable and elastic. We also have a complex physics simulation that greatly enhances the playback, with all the parameters of a driving vehicle, including a gearbox model, a virtual clutch and a layered load/offload mixing system.
DS: And you recently announced your integration with Wwise?.
Adam Boyd & John Twigg: REV is deeply integrated with Wwise, and is already shipping with Wwise as of version 2013.2. The REV Authoring tool is easily launched from within Wwise, and the content parameters can all be tweaked live within the game. Additionally, while offline, a realistic simulation of the engine can be generated within Wwise to hear the audio experience in context. Last year we spent some time in Montreal working directly with the AudioKinetic team to ensure that the tools were working cohesively. We also offer a great selection of inexpensive, ready-made REV engine models for developers who don’t have the time or resources to go out and record their own cars, and these are all available to check out on our Wwise partner page. Audiokinetic is a great partner, and we feel this is a really strong symbiotic relationship.
DS: Having had a chance to play around with a demo of REV I was impressed with it’s ease of use, and the speed of the turn around, something Adam and John have worked hard to achieve.
Adam Boyd & John Twigg: One of the biggest strengths with REV’s authoring tool is the turn-around time in analysing, tracking and preparing game-ready content. The ‘traditional’ approach, which we feel is an inferior approach, is to create a stack of crossfaded loops, and pitch bend them to simulate RPM changes. This approach is time consuming, difficult, and in the end, creates a poor car engine experience. The turn-around on a loop model can be 3 to 4 days. With REV, it’s on the order of 10 minutes. Game audio teams these days are burdened with a huge number of tasks like speech, guns, environmental audio, music, collisions, Foley… the list goes on. To add cars on top of this – which are notoriously difficult to do well – is the final straw. We feel that with REV they can just “tick the box” for engine sounds, and focus on the rest of the experience.
DS: With some of the technicalities out of the way I wanted to know what is it in the design and implementation of these engine sounds that really makes them sing for the gamer?
Stephen Baysted: For Project CARS it’s fair to say that Greg and I (and the highly talented audio programmers without whom very little would be possible) have broken much new ground in terms of the detail and complexity of our models. We’re about 4 months from completion now so there are a few aspects of the sound modelling that will be refined and we do still have some new features to add. Our approach is, as always, to give the player the most realistic sonic experience possible within the limits of current technology. Whether we have achieved that by game release is another question! In terms of design, we’ve captured recordings from a very large number of cars in a greater level of detail and quality than previously; the number of crossfading layers is far higher than earlier projects; and our audio programmers have been implementing some very creative ways of making the vehicle audio interact with the track environments.
Greg Hill: It’s the marriage between sound and physics. If the physics are awesome there is no reason the sounds shouldn’t be awesome too. There should be a sonic response to every physical event and it should be dynamic so the variations in sound should be realistic. Think of the sonic model as a musical instrument and the physics as the player. You can have the most beautiful sounding instrument but it’ll always sound like crap in the hands of a hack!
I think iRacing’s physics is pretty darn good lately, in fact the alpha build I have right now the tyre model is unbelievably good and you can hear the thresholds of grip from rolling to scrubbing tyres and then total grip loss. The engine load and throttle input is also great, you can hear the engine change with the smallest variation from your foot and perceive the weight of the car as the engine loads change according to the conditions like bumps, inclines and even the transmission whine wobbles and lurches with the drive train stresses.
Nick Wiswell: The key for a simulation game is how the audio ties into the physics system, and making sure that both are accurately representing what they need to.
We use hundreds of physics based parameters to drive the audio system in Forza games, including RPM (independently for engine, turbo, supercharger, transmission etc.), throttle position, engine load, torque, boost pressure, clutch position, gear selection and many others. When audio is reacting correctly to all of these inputs, the car feels more believable to the player.
For example, let’s look at the exhaust sound, which is just one part of the overall sound of the car. Here are some of the questions we need to answer:
• How does the sound character shift between on and off throttle?
• How does positive or negative engine load impact this?
• Does it sound different in different gears, if so why?
• Does turbo/supercharger boost pressure change the sound?
• Does the speed at which the throttle is applied change the sound?
• Does this car “burble” or “backfire” when decelerating?
• If it does, how often and what physics parameters should drive this?
• Does the sound of the car change with speed?
This should give you some idea as to the some of the thought process that go into creating just one sound component within the car.
DS: On Forza 5 the guys at Turn 10 took their thought processes a stage further with a visit to Skywalker Sound.
Nick Wiswell: As a self-confessed Star Wars geek it was one of the best days ever, but it was a huge learning experience and a very worthwhile exercise. We approached them early in the concept phase of Forza Motorsport 5, and asked them to create a soundscape for a video of one lap of the Alps track in Forza Motorsport 4 as if it were an action movie. We gave them a few weeks to develop ideas, then my Audio Lead Chase Combs and I flew down to Skywalker Ranch to listen to what they had done, but also to talk about why they had done it, as for me that was the most important part of this process.
The key takeaways from this were that their approach to sound design and mixing were very different to ours, as in a movie you do not have to consider the player feedback that audio can give you. For example there was one section where the engine audio was totally inaudible, but we cannot do this if the player is using manual gears as they need the audio to help them know when to shift. But there were several other approaches that they took that we did find really useful and all of these feature at some level in Forza Motorsport 5 and will continue to drive our creative process for audio as we move forward onto new titles. These include:
• Shifting the focus of the mix to the thing that is most important for the player to hear at any given moment
• Using the surround field in creative ways to ensure everything has a place in mix
• Using the players car as our leading actor, and using the center channel in a similar way to how people approach dialog in a movie
• When you don’t have scenes to create sonic changes (usually in the game the player uses a fixed camera), make scenes using what you can see to keep things interesting.
DS: During my research I couldn’t help but notice how gamer’s react to the quality of the engine sounds on display. In my travels I encountered a variety of positive and negative comments, both equally passionate in their opinion. I asked the contributors if theirs a conceptual gap between developers delivering authentic engine sounds and player’s expectations of these sounds?
Stephen Baysted: Good question – the answer could turn into an essay! I’m not sure it’s a conceptual gap per se but there are many interwoven issues here and the topic is a veritable can of worms! One interesting and fairly recent issue is the advent of You Tube videos, and in particular the low quality on-board videos of cars captured on the ubiquitous Go Pro cameras or mobile phones. For many gamers, these videos are and remain their only point of reference for the real car. Clearly, a carefully recorded car with maybe 8 DPA mics (positioned in the engine bay, intakes, exhaust, turbo intake, driver’s seat) and captured through high end pre amps and meticulously post processed will sound a little different! :) Try convincing some gamers which is the more realistic recording!
Another practical issue is of course trying to recreate the sheer sonic brutality of the real racing cars. Their elevated SPL, their wide frequency ranges and in particular 50hz and below into the subsonic realm means that one simply cannot hope to reproduce this ‘real world’ experience authentically on the average gamer’s computer speakers in acoustically compromised listening environments.
Yet another problem is attempting to recreate what the real driver hears (with his fireproof balaclava, ear plugs and crash helmet on). We have a mode in the game called ‘Helmet Cam’ where the player’s viewpoint is from the driver’s eyes and this perspective changes with G-forces and other physical inputs. The sound in this mode also changes – principally through comb and low-pass filtering – but many players do not find this view sonically exciting even though it is actually very accurate. But accuracy is not always desirable in a gaming context, even in a painstakingly simulated one. Because the player is obviously not really driving, one is forced to give sonic clues to help them drive – one of these is making the tyre sounds much louder than they are in real life so the player understands the limits of grip and traction.
Nick Wiswell: This is one of the biggest parts of my job, trying to understand different people’s perception of authenticity. When creating the sound of car we need to take in the several factors:
• How our recording of the car sounds
• If one of our team recorded the car how they remember the car sounding on the day
• Our players potential interactions with the car in real life
• Has this car featured in a motion picture or TV show and how did it sound
• What does the car sound like on internet videos
A combination of all these factors will be used to determine how people will perceive the car to sound, which in many cases may not be exactly the same as what we recorded. A car is also a very complex sounding thing, with many layers of different sounds creating the overall sound you hear. These layers of sound change continuously based on what the car is doing and where you are positioned relative to it.
Creating car sounds for linear media requires you to record the car doing the same thing it does in the scene from the position of the camera, as the scene will never change once it’s completed. In a game we have to try and capture the sounds of the car doing everything it can ever possibly do from every position possible on the track, then deconstruct that in a way that allows you to piece it back together again based on what you are doing and where you are in the game at any given moment. That’s a difficult thing to do with a fairly simple sounding object, so doing that with something as complex as a car is very tough if you want to get ever single permutation of sound source (engine, intake, exhaust, transmission, turbo, supercharger etc.), physics input (rpm, throttle, load, gear etc.) and position relative to the camera and the world around you correct.
With the development of every game we are learning new techniques to make our cars sound better, and the response from our Forza community is always welcomed and informative as we move forwards.
Greg Hill: Most players are enthusiasts and want a realistic experience – as do I. Sometimes they will blame the sound department and ultimately this could affect the sales of the title. Some players aren’t aware of the limitations of the technology and just blindly want it and want it now and moan a carry on in forums. It can be a real challenge to get the very best out of an audio system if it’s lacking in maturity. You can be the best sound designer in the world but if the system is substandard or the developer/publisher is going all cheap on the audio they ultimately pay for it in the end with bad reviews and sales. The main characters in a racing title are the vehicles; they need a voice – a darn good voice!
Adam Boyd & John Twigg: Racing games in particular inspire heated debate about the verisimilitude of their engine sounds. It’s a subjective experience, and sometimes a “realistic” sound isn’t the right sound depending on the game. However, the ability to capture the unique character of a vehicle and reproduce it in a way that retains all the aggression and power of the original recordings without significant degradation or introduction of pitch artifacts is a valuable tool for developers. Most sound designers we know who specialize in cars are very passionate about their work, and spend many hours sourcing the right car (and convincing the owner of that car to let them record it) and many more hours mixing and mastering those recordings. REV is the best tool to maintain the consistency and authenticity of those recordings right through to the game implementation phase.
Many thanks to all the contributors for their time and input.
Greg Hill – Greg has more than 18 years and 80 titles under his belt. He specialises in vehicle sound recording and design.
Stephen Baysted – Stephen is a composer, sound designer and auido director with a varied credit list including AAA titles, film scores and high profile ad campaigns.
Adam Boyd & John Twigg – Adam is a sound designer with credits including Blur and Modern Warfare 3. John Twigg is a software engineer who has worked at Piranha Games and Criterion Games. Together they formed Crankcase Audio and developed REV.
Nick Wiswell – Nick is the Audio Creative Director at Turn 10 Games and has worked on Forza Motorsport 4 & 5.