So what are your options? Buy a sample library of airplanes? Great, most of them are very good quality, no background noise or disturbances. The problem with this is that they have all been used and may not have the samples that fit your requirements. Another option is to go guerilla and stand at the edge of the airfield with the plane spotters. This is a good day out but you can come across problems with security and the public. Fred Pearson captured some great helicopter flybys like this but, some of the recordings were made unusable because of plane spotters making noise. To read about it go here: http://fredpearson.wordpress.com/2011/10/22/helicopter-recording/

Another major problem of recorded audio is effects such as doppler and reverb. It would be difficult to record a plane take off without a doppler effect over the airplanes sound. You might want to use that sample but it wouldn’t fit in with the camera perspective.

There is an alternative option: synthesizing the sound of the airplane procedurally. Andy Farnell’s ‘Designing Sound’ shows a basic model for creating a jet airplane. Let me walk you through this procedural model and talk about how it works.

The main noise emitting components of a jet airplane are the engines. This can again be broken up into two parts; the turbine and the burner. The burner is the turbulent flame burning sound, this is simply made up of banded distorted noise. As you can see from the model the control signal (which ranges from 0 – 1),  is first squared and multiplied to give a range from 0 – 150. Then the signal controls a filter which acts upon noise centred around 2000 Hz, creating a shift in bottom end frequencies. A high pass filter is used to safeguard against any low frequencies which would cause audible fragmentation. This signal is amplified and distorted to induce harmonics. Another variable filter is used to add dynamics in to the sound, with a low control signal not much of the noise passes, with mid control signals there is lots distortion, but with full control signal the filter centres at a higher frequency than the noise so there is less distortion. I have added more low frequencies back into the signal using a two stage low pass filter and some gain. This adds some real weight and size to the sound.

The turbine creates a loud ‘whirring’ sound. In this model it is made up of individual sine waves added together. From spectral analysis of an airplane sound certain frequencies can be picked out, their relationship is more important than their actual frequency. So the 5 main frequencies from the spectral analysis are put into a message and unpacked into 5 signal generators, their frequencies are then modulated according to the engine speed. I have given faders to control the level of each harmonic in the turbine, this allows more control and modelling to taste. Finally the sound is then overdriven to induce more harmonics.

The turbine and burner are put together into a signal engine unit. The control signal is made by a fader ranging between 0 – 1. Using a low pass filter on this signal slows the movement of the signal. Some crude level adjustments are done on the turbine (each could easily be controlled via a fader). I have added four engines together with slightly different low pass filters, this makes their control signals to vary ever so slightly during movement, creating more separation. Finally the sounds are panned across the sound stage, low pass filtered and gained down.

In conclusion this model is simple and effective. Using this in a small flight simulator would free up a lot of memory that would be taken up by sampled audio. The model has a few parameters that can be controlled and used to add variation to the sound, causing less repetition. Compared to sampled audio I think this model sounds great! A few other effects and parameters need adding such as the doppler effect and environmental effects.

So here is what you’ve all been waiting for…. audio! WARNING these are normalized to -3dB and are pretty loud, watch out!

Andy Farnell’s ‘Designing Sound’ covers numberous models for procedural audio. After experimenting with the wind generation I wanted to implement the thunder model because of its relationship to weather systems. In future posts I will cover other aspects of synthetically generated weather such as rain and various objects affected by wind. Hopefully, I’ll be able to integrate all of these systems together.

Thunder, created synthetically, is quite difficult to model mainly because it’s sound is largely dependant on environmental factors. The surrounding areas affect how the sound is reflected, delayed and filtered. In a game environment it would be very CPU intensive to model the sound (and environment) correctly in real-time. Although this system can be done quite crudely because of the masking and blending of the sound. In the future, it would be great to hear a thunder system in a game that would also model the effects of the environment, either using ray tracing in real-time, or using a hybrid reverb system. Wwise’s latest update now includes a hybrid reverb: “Hybrid Reverb allows the user to import an impulse file, extract the early reflections, and separately control the reverb tail.” (Audiokinetic, 2012). This kind of system would allow a pre-rendered ray-traced impulse response to be used which could then be manipulated to appear real-time.

So on to the thunder model. Again this designed from Andy Farnell’s book ‘Designing Sound’, I recommend anyone interested in procedural audio or game sound to go buy it and read read read. You’ll find a full walk through on how to design all of the patch and many, many other models.

This model uses a few abstractions and consists of 6 main parts. First part being the ‘pd strike-sound’, this patch uses 4 instances of the same abstraction as well as a pattern generator for the initial strike sound. The pattern generator creates decreasing values that separate further over time using counters and delays. The 4 strike sounds are then triggered to form a clatter at the beginning and start to separate. A lightning bolt forms this because of the distances (causing delays) involved in its propagation and air absorption (decreasing values). The sound generated inside each strike are a pair of band passed filters, noise and an envelope. The first band pass filter can range from 100 Hz to 1300 Hz, the second filter replicates the octave below. The strikes then become longer as the input float value is lower, causing a nice natural progression through the sound.

‘pd rumble’ is the detailed sound after the initial strike of the shockwave that propagate from the lightning bolt at various delays. Farnell discusses that this is the most difficult part to model, and provides references to physical models from other researchers, but states they are computationally expensive. So a crude approximation utilised for real-time has to be used. The final result is a “randomly rising and falling ramp which is then shaped to produce parabolic pulses with a controllable density” (Farnell, 2010). Two noise generators are low passed, one uses only the positive part which drives a phasor at around 10 – 20 Hz. Using a samplehold~ the first low passed noise freezes when the phasor falls causing random step heights and intervals to form a stepped waveform. rpole~ then produces a jagged triangle waveform, then a parabolic shaper softens to waveform into a rumble. The model has an envelope applied to fade the rumble out. The abstraction also allows for density envelope so the sound develops into a more sinusoidal waveform over time.

‘pd afterimage’ synthesizes some of the effects causes by the environment. The environment effects the sound through the reflections from numberous surfaces, these are also filtered and modulated because of the surfaces properties. Initially the abstraction has a delay of 200 ms and uses noise sources which modulate each other. The model cleverly uses a large gain boost on low passed noise and then clipping to cause a stuttering and gating effect on the signal. The low pass filter also implements modulation on the cut-off frequency (which is in the form of an ADSR), this slowly reduces the frequency causing less stuttering. Finally band pass filtering this around 300 Hz  imitates the reflections from various buildings, ultimately causing a ‘clatter’ effect.

‘pd deep’ creates the low-frequency content that give the sound some real body, this is mainly heard during distant thunder storms due to air absorption attenuating higher frequencies. How this works? Simple really; white noise, low pass filtered below 80 Hz, then high pass filtered (to reduce distortion and any DC offset), expanding the dynamic range, then clipping the signal to induce harmonics. Finally, this gets low pass filtered at 80 Hz once again to reduce some of the high frequency. This then needs an envelope; a delay of 1 second before the envelope starts, the control signal then rises to 1 in 2000 ms and falls to 0 in 10 seconds.

‘pd box of delays’ is simply a multi-tap delay. Each refection will have its own characteristic tone and random panning. This be replaced with an impulse response or using ray tracing in an environment to create the reverb. Although all of these processes are CPU expensive because they are essentially a collection of individual delays. This abstraction uses a random number to set the delay time. If the delay time is longer then the panning is further to the right and they are band passed to a lower frequency, if the delay is short the reflection is panned to the left with a higher band pass frequency. Creating more instances of this abstraction makes the reverb more dense, but more CPU intensive.

The distance abstraction simply comprises of low pass filters and a comb filter for the ground reflections. I’ve also added faders to each model so that any mix of thunder can be created. All of the filters and delays can also be changed to make a vast number of thunder sounds.

Here are some quick audio examples of the thunder model. These each have slightly different mixes between each abstraction. None of the delays or envelopes have been changed, but this is also easy to do. Please bear in mind this samples are only mono so you lose some of the sonic quality, I’m currently working on a stereo recorder within PureData.

In conclusion this is another great procedural model, it would be great to see this implemented within another weather system along with rain, wind and rolling objects. In contrast to sampled thunder the sonic qualities are not as appealing as the sampled audio, especially when compared to a sample library such as Frank Bry’s ‘Thunderstorms‘. But then again, as a full procedural weather model this could take out a lot of sampled data and audio channels within a game, with a reasonably small computational cost (excluding modelling the reverb).

References:

Audiokinetic (2012). Wwise 2012.1. available: http://www.audiokinetic.com/en/about/news/219-2012-1.

Farnell, Andy (2010). Designing Sound.

One of the procedural models that I have found fascinating has been wind generation. This is an important aspect of games as it adds to the ambience, feel and can help localization. Through traditional methods wind would be produced using linear loops or samples stung together dependant on an algorithm or variable. In contrast to procedural audio, the sampled method uses more data storage and has less variation available. So this seems like a great place to start implementing procedural models within games; this has already begun by the release of ‘AudioWeather‘ by AudioGaming. Here is the procedural model from within Designing Sound:

I’ll take you through a quick run-down of how this works, then in a following post I’ll provide some audio clips.

The model starts at the patch ‘pd windspeed’. This contains two other patches called ‘pd gust’ and ‘pd squall’, together these simply make the control signal of the wind sound. The wind speed is not linear which causes the control signal to be slow and steady at low speeds but it grows at a squares law creating heavier gusts of wind above half speed. The ‘pd squall’ utilities any signal in the last 20%, so during strong gusts the sound becomes evermore chaotic.

‘Swept Noise’ – This uses the control signal to modulate the amplitude of band-passed white noise. As the control signal increases the band passed signal increases. rzero~ has also been implemented which creates a slight filter over the broadband noise, this acts as a variable filter changing according to the control signal.

‘Whistling Wires’ – Each unit is a narrow-band variable filter whose centre frequency is a function of the windspeed. So the stronger the wind the higher the frequency. Their amplitude follows a square law so their volumes are low at low speeds. Adding 0.12 offsets the point that the sound starts, imitating a real effect. The effect of panning and variable delays creates a field of view so that the sound starts at the left, and as the wind blows across it excites the wire in the right.

‘Tree Leaves’ – This is the whole models weakest aspect but in essence it uses delays the control signal by two seconds (to simulate a similar effect of leave being pushed by the wind). The control signal then gets reduced and filtered so that with a lower signal there are less noise spikes let through. The stronger the signal the denser the sound is.

‘Howling Wind’ – This unit creates delayed ranges in which the control signal will pass through, causing the resonances to occur when they reach a certain windspeed. This also utilises amplitude modulation to create side bands of the resonances. Ideally this model should be attached to objects within the game world to create an immersive ambience.

Here are some audio clips of the model:

How does this work as a procedural model? In all honesty it doesn’t output the highest quality sound, yet in a game environment I think this would be pretty convincing. The sound reacts as wind does and has very similar characteristics. The advantage of using this type of model over sampled data is that is completely variable. Each modelled part can be abstracted and applied to in-game objects and environment. This means using the same model can be used for indoors (more howling, no leaves and some more filtering), in a field (less howling, more leaves and broadband noise.

Overall this is a simple and effective model and I will continue to work and improve the model to make it more efficient, realistic and variable. Key area that need work are: The tree leaves, more variation and control, rolling and blown objects, as well as overall filters and sound shaping.

Keyboard Hero:

This project was to get down the key principles of scripting with Actionscript 3.0. If you have ever played Guitar Hero you’ll know how it works. The has four keys, when you dots or ‘notes’ hit the targets press the corresponding keys. The better you are the better your score will be. Throughout the development process of this game I learnt many features of Flash, gained a basic grasp of scripting and this has enabled me to understand how code works written in other scripting languages.

Music Sequencer:

To push my knowledge of Actionscript 3.0 further I decided to design a music sequencer. Now this task required much more programming and intensive logical functions. The main features I learnt from this project was the use of timer functions, arrays and creating interfaces to interact with the sound.

So I bring to you a sampler of the ARROWHEAD AUDIO SFX ‘SCRAPES’ Sample pack. This will be released this summer once the sample pack has been finalised.

]]> http://mattmeachem.wordpress.com/2011/07/31/rusty-metal-scrapes/feed/ 0 mattmeachem Scrapes Arrowhead Audio http://mattmeachem.wordpress.com/2011/06/25/bayonet-sfx/ http://mattmeachem.wordpress.com/2011/06/25/bayonet-sfx/#comments Sat, 25 Jun 2011 20:59:18 +0000 mattmeachem http://mattmeachem.wordpress.com/?p=73 ]]> During my recent Sound Effects recording adventures I came across a particular item which I thought could, in the future, become very useful. This was a short Japanese Bayonet. The casing was very rusty but the blade was in very good condition, albeit very oily. Here is a picture of the blade and casing…

When I found this (rumaging through my Grandparents garage), I immediatly thought this could make, or be part of, a very useful sample pack. Straight away I started to tap, bang, slide, scrape, shake and draw the sword. Consiquently these are just some of the samples recorded from this bayonet:

More samples are to be added to this pack such as higher intensity hits. The next items on the SFX list are hammers and small axes. These will all combine to create a new Arrowhead Audio Sample pack

I came across the bathroom cupbard, which makes a long high pitched squeak. I didn’t previously noticed it as I have heard it day in day out for the last 15 years. So put my recorder inside and started to ‘play’ the door. These are my results:

When experimenting with the sounds during editing I halved the sample rate causing the pitch to shift and layered sounds up. I came up with a sound that is of similar nature to that of an air-raid siren. Great fun is to be had in future projects by layering this sounds into various designs.

Two short video clips were chosen for my portfolio. One was an animation from Pixars’ shorts of the characters from ‘Monsters Inc.’ The second was a game cinematic from the game ‘Borderlands’. These were chosen because of the opportunities they offered. They both needed a lot of work in many different areas of post-production, including; ADR, field recording, sounds design and surround mixing.

Here are the finial clips:

Mike’s New Car

Borderlands Introduction

Any feedback would be greatly appreciated.

These have been my first pieces of portfolio work which have given me opportunities to work on other Audio Post-Production such as for Beyond The Bar Productions.

The recording contrains the old locomotive, with around 8 carriages, pass by.The H4n recorder was placed around 1.5 meters from the rails, pointing directly where the wheels are. This was to capture the ‘clunks’ and screeches of the train as it goes by, The train was surprisingly quiet and can be heard to approach very fast, maybe next time positioning the recorder a few more meters away from the train to capture a wider field. So here is the sound:

I’ll be planning to go up there again and record some more trains passing by but at different distances, maybe with a few different microphones and positions. An interesting test might be trying a contact microphone on the bottom of the rails or on the floor, this would capture much more rumble and could produce very good results.

I only used the onboard stereo microphones as I didn’t have a dynamic microphone at hand. The microphones were positioned pointing up towards the end of the barrel, placed about knee height off the floor under the gun. I didn’t have time to compare positions for recording these guns, but this will definately be a good venture when I have more channels to play with or I am in a more controlled environment.

The shooting range was outdoors, in the countryside, and there was a main road  within 100 metres of the range. This could have been a major problem on the recordings but as the ranges are situated within the hills of the countryside, most of the time all sides of the recorder were shielded from direct background noise.

The main problem I encountered with this recording session was how to record a clean signal but with enough sound. I was not sure whether it would be best to record the loud transients without clipping, obviously this would technically be best because it causes the least noise. The other possibility was to let the recorder clip or use the onboard limiter, causing distortion, but the recorder would pick up much more of the lower level detail. Which sounds better will be found after listening to the sound files.

Overall, this was a great day out testing the Zoom H4n. The next time I go shooting and record some of the shots I will be taking a few microphone stands and a dynamic microphone. This is so I can get the recorder up close to the source of the sound without the risk of damaging the onboard microphones and provide a slightly different perspective for the sound. And re-reading over Rick Veirs’ ‘Sound Effect Bible’ before the sessions too, so that I can pick up on any extra tips that he might provide.