96808_P18_20_Noise:FEATURE/TECHNICAL PAGE 23/10/08 09:12 Page 18 Dialogue Noise Suppression TV walls, moving lights, and the fashion for hard reflective set designs has made capturing clean dialogue a serious challenge in modern television. Digital Noise Suppression offers a The CEDAR practical solution as Peter Best DSN10000 dialogue explains. noise suppressor he modern broadcast environment The use of dynamics new, more precise can be a source of many types of processors such as noise gates technology possible: T noise and unwanted sounds. Some and full-spectrum expanders was no spectral subtraction. This of this, such as room ambience and noise better, resulting in unnatural break-up of technique splits the introduced by the re-use of existing noisy the signal or unpleasant noise pumping. spectrum into numerous material, is unavoidable. Others, such as In the 1960s, the multi-band expander bands (typically 512 or 1,024) and uses a those caused by undiagnosed electrical was developed. An early version of this split noise fingerprint to determine whether the faults and poorly positioned equipment the spectrum into four bands and, if the amplitude in each has dropped below the can – with care – be cured at source. But signal level in any band dropped below a threshold below which gain reduction is many noises cannot be eliminated by careful user-defined threshold, the amplitude of applied. Although expansion and spectral microphone positioning, intelligent that band was progressively suppressed subtraction are usually viewed as separate placement of equipment and good according to a simple gain ratio. This technologies, one is merely an extension of maintenance, and it is here that broadcasters technique worked on the assumption that the other, albeit with different operational must turn to signal processing to suppress or, signals with amplitudes above the threshold characteristics and different side effects. At when possible, eliminate the offending contained wanted audio, while those with one extreme, the single-band expander sounds. amplitudes below the threshold were just generates noise pumping; at the other, the noise. Conceptually simple, this was high 1,024-band spectral subtractive dehisser can Evolution of Noise Reduction technology for its time, and a product based generate artefacts usually referred to as Traditional methods of noise reduction on this idea – Dolby Lab’s ‘Cat 43’ – proved twitter, gurgling, and even ‘space monkeys,’ included the use of low-pass filters to to be surprisingly effective. Furthermore, it depending upon the nature of the input and reduce high-frequency hiss and high-pass was simple to use, with a physical control the processing applied. filters to eliminate rumble. Unfortunately, surface that was both immediate and Spectral subtractive systems can be very depending upon the signal spectrum, these intuitive, so many of these units are still in precise about what noise they remove and affected the wanted signal as much as the use today. can produce remarkable results, but they noise and nowadays this may be The advent of large amounts of are not always simple to set up optimally. unacceptable to a high quality broadcaster. computing power in the 1980s made a Furthermore, they assume that the noise content is constant over a reasonable timescale, and this is not always the case, especially when microphones are being moved or when faders are being raised and lowered. For broadcasters, there is an even more serious limitation to spectral subtraction. The algorithm uses mathematical techniques that require significant signal buffers, resulting in a few video frames of latency. Furthermore, with very few exceptions, spectral subtractive algorithms are implemented on computer platforms that add further delay in their I/O stages. While acceptable when working off-line, this latency often renders spectral subtraction inappropriate for use when the sound and picture have to remain The Dolby 430 Series Background Noise Suppressor (based on DolbySR) synchronised, especially when broadcasting replaced the Cat 43 unit (based on Dolby A) in 1992 live to air. 18 LINE UP Nov/Dec 2008 96808_P18_20_Noise:FEATURE/TECHNICAL PAGE 22/10/08 12:06 Page 19 Dialogue Noise Suppression “If there’s a downside, it’s that In the 1990s, many production houses and broadcasters had begun the transition from analogue to digital audio but, due to the lighting directors now take limitations of computer-based techniques, they were (by and large) still using analogue advantage of noise expanders as their primary noise reduction systems. Clearly, there was a requirement suppression.” for a digital solution that combined the simplicity of analogue units with a modern, Noise Suppression, or simply ‘DNS,’ Big Brother is Listening digital algorithm that (as much as possible) combined elements of existing analogue Oliver France of Oliver France Sound was avoided the twin evils of noise pumping and digital processes, and had a negligible an early adopter of the CEDAR system: and twittering in the de-noised signal. The latency of just five samples (approximately “I was engaged for the first series of Big solution also needed to exhibit near-zero 0.1ms at 48kHz, or 1/400th of a frame at Brother in 2000 or thereabouts, and the latency, such that a number of units could 25fps). directors were constantly seeking new be daisy-chained with no perceivable loss of DNS made its first public appearance in ‘looks’, most of which generated large lip-sync. 2000 in the form of the CEDAR DNS1000 amounts of audio noise – such as big Several algorithms were proposed and Dialogue Noise Suppressor. It was not the screens and lighting with built-in cooling at least one noise filtering product appeared most powerful noise reduction system fans. On many occasions Davina McCall was in the mid-90s, the GML 9550 Digital Noise available – time spent on a spectral surrounded by screens and lights and the Filter. Meanwhile, in the UK, noise subtractive system could often yield noise was challenging the programme audio reduction specialists CEDAR Audio were superior results – but it was simple to use, to the extent that it was not fit for experimenting with processes based on an benign in its generation of side-effects, and broadcast. The problem was to reduce the intermediate number of bands. These were its near-zero latency meant that it could be noise but still hear what we wanted to hear. designed to suppress background noise in used in real-time during broadcast. I had heard of the CEDAR box, and this signals with a reasonable signal to noise Furthermore, it was able to suppress a wide seemed an ideal opportunity to see what it ratio, and (as far as possible) to minimise range of unwanted sounds including could do. We brought in a DNS1000 and a unwanted side effects. The resulting general background noise, babble, lighting Sonifex Redbox ADDA and the results were algorithm, which they called Dialogue buzz, and camera (shutter) noise. NTG-3 THE EXTREME LOCATION MICROPHONE LINE UP Nov/Dec 2008 19 96808_P18_20_Noise:FEATURE/TECHNICAL PAGE 22/10/08 12:06 Page 20 amazing. We managed to filter out the “In general, we apply the DNS1000 broadcasts. This is essentially a 10-storey majority of the noise without losing the across the studio mic group so that it is hole and, in the General Election broadcast quality of the source material – Davina’s cleaning the signals from all the of 2001, the combination of a long reverb voice. Fortunately, once the soundstage was microphones simultaneously, and the time and huge, fan-cooled screens could rearranged the noise was pretty constant, so output is then compressed before being really have stuffed us up. To cure this, we we could set up the DNS1000 and leave it presented to the broadcast chain. However, used a DNS1000 in a novel way, with its left to do the job.” the way we use it will depend upon the channel dedicated to the microphone of In 2006, Oliver moved from Big situation in which we find ourselves and, one of the main presenters, and its right Brother to Deal or No Deal, where the because the noise changes during the channel dedicated to the other. The noise problems were somewhat different. “The transmission, the sound supervisor will that each presenter was experiencing was studio is built into a warehouse-type unit in often tweak the settings on the fly. Also, we quite different, but when set up in this way Bristol, and unfortunately you can hear have found that if there is a lot of low the DNS1000 coped admirably.” everything going on around it. There are frequency noise, we can attenuate this with Noise suppression is required in post- also a lot of moving lights and a large EQ before presenting the signal to the DNS. production at least as often as in broadcast. screen, all generating noise. However, The noise suppressor will then do an even In this environment, additional unlike Big Brother, it’s possible to have up better job with less risk of pumping.” considerations become important. These to 22 gooseneck mics open as well as Noel include the ability to automate the noise Edmonds’ radio mic, so the noise content is Excess Reverberation reduction process, synchronisation to constantly changing. At first, we used a Another problem encountered with programme content, and integration with DNS1000 in post-production, but after the increasing regularity is that of excessive commonly used workstations. It may also first few shows we moved it to recording, reverberation, as Nick Ashton, Sound Team be desirable to clean multiple channels placed on an insert of the talking group so Leader at ITN News, explains: “We produce simultaneously, either when running that it did not process the music or studio the news broadcasts for ITV, Channel 4 and multiple channels of dialogue, music and applause.