CineClean Pro A Noise Reduction Application Proposal for Final Cut Pro by Matt Ozer
In partial fulfillment of the requirement for the Bachelor of Science degree in Audio Production at the Art Institute of California at San Francisco
CineClean Pro A Noise Reduction Application Proposal for Final Cut Pro by Matt Ozer
Abstract In the film industry, due to limited resources and inadequate environmental conditions, poor or unusable location sound is often recorded. As new technologies develop, filmmakers now have access to tools once only accessible within post-production facilities. This research proposes the development of an application that will improve and enhance the fidelity of location sound recordings by implementing a noise reduction application that is easily accessible from within the Final Cut Pro environment.
Chapter I â€“ Introduction 1.1 What is noise? 1.1.1 Overview In the film industry, due to limited resources and inadequate environmental conditions, poor or unusable location sound is often recorded. As new technologies develop, filmmakers now have access to tools once only accessible within post-production facilities. The new standard of High Definition (HD) requires a new set of audio tools in order to preserve and enhance this high definition in sound reproduction. This research presents a proposal for the development of a noise reduction application called CineClean Pro, as well as examining the history of noise and several comparisons of modern noise reduction techniques including Bayesian estimation theory, Hidden Markov models and Gaussian process. By addressing the above topics, this research proposes to develop an application that will improve and enhance the fidelity of location sound recordings by implementing a noise reduction application that is easily accessible from within the Final Cut Pro environment.
1.1.2 Definition of Noise There are several definitions of noise. We might consider noise as the hiss when listening to a radio station or the flickering snow when trying to watch scrambled pay per view shows. Or, perhaps it is the grain on an old film or the background chatter of people in a restaurant when we are trying to hear tonightâ€™s specials. Noise could be the boom pole in an otherwise academy award winning shot; it is an entity that does not belong there. In essence, noise is an unwanted signal. A signal is anything that conveys information, such as an electrical pulse, a chemical pattern, or the acoustical roar of an audience. Noise can be physical, chemical, biological, or even conceptual. Noise is also subjective, as it impacts how humans discern and evaluate desirable signals from undesirable signals. Furthermore, noise not only can obstruct a signal, but can also interfere randomly. This interference varies from person to person, as Bart Kosko notes: A meteorologist or photographer may wait for days to see lightning strike an apartment building. Both count the lightning discharge as a good signal. But the doorman or the resident who is surfing the Internet may recognize the same lightning discharge as a noise impulse.1 The same energy can be perceived as signal to one person and as noise to another. Within everyday occurrences, noise is distracting. Our focus can easily be disrupted by unwanted clutter and sounds from our surrounding environments.
Kosko, Bart. Noise.
1.1.3 Why noise? If you live in a city, then traffic noise, car alarms, police sirens and cell phones ringing can interfere with your daily life. This is especially true when these sounds interfere with desirable signals, such as speech. Living in a noisy environment can not only cause hearing loss in millions of people, but also stress. Almost everyone has suffered sleep loss at some point because of noise from traffic, construction, or emergency sirens. A recent report published by the World Health Organization (WHO) and the European Commission's Joint Research Centre states that “Steady exposure to ‘noise pollution’ can also lead to noise induced hearing loss and several other health problems.”2 It is likely that this problem will continue to pervade our living environments in the future. The goal of this research is to develop a noise reduction application that will hopefully combat the presence of these unavoidable interferences from location recordings. It offers a creative solution that will reduce undesirable artifacts in location sound recordings, thus creating a more enjoyable presentation and experience.
Peterkin, Tom. “Noise Pollution Map Warns of Health Risks.”
1.2 History of Noise 1.2.1 Noise in Industry and Art With the development of mass transportation systems in urban areas, noise pollution became a reoccurring problem, and as such, many solutions were considered. In the 1880’s Mary Walton first invented a noise reduction system for elevated railroads in New York City (see fig. 1). In order to reduce noise; she invented a “sound-dampening system that would cradle the track. It was essentially a wooden box lined with cotton and then filled with sand.”3
Fig. 1. Elevated Railway: one of the first major noise pollution problems in urban areas. Project for a Parisian Elevated Railway; (Scientific American: Munn & Co., 1886). Print.
Within the arts, noise had different associations. Luigi Russolo, a painter who was part of the Italian “Futurist” movement at the turn of the century, is arguably the first noise sound artist. Russolo published his manifesto L’arte dei Rumori (The Art of Noises) in 1913.
Bells, Mary. “Mary Walton”
Within this, he proposed a new philosophy of creating music, which included noise (see fig. 2). [Russolo] envisioned huge musical works made from mechanical sounds of everyday life such as revving motors, valves opening and closing, moving pistons, power saws and streetcars, as well as other types of noises, like cracking whips and sounds from flags flapping in the wind, etc. To capture the quality of these sounds, Russolo invented â€œintonarumoriâ€? (i.e. noise instruments) that can mechanically produce many timbres over a range of pitches.4
Fig. 2. Luigi Russolo in 1913 with his mechanical orchestra. Intanarumori. The Art of Noise. (Great Bear Pamphlet: Something Else Press, 1967) 13. Print.
American composer, artist, and philosopher John Cage was undoubtedly the central figure in the redefinition of sound from the 1950s to the present. While artists working in the early twentieth century generally created harsh noises of industry and machinery, John Cage listened for the subtle harmonies that were randomly generated in both natural and acoustically built environments.
The Art of Noise. (Great Bear Pamphlet: Something Else Press, 1967).
Cage’s most notorious piece, 4’ 33” (1952), was composed using silence. This was an important milestone for many other composers. In this piece; A performer sits at a piano for four minutes and thirty-three seconds without producing a sound, simply turning the pages of the score and closing and opening the piano lid to indicate the three ‘movements’ of the piece. Chance-determined ambient sound (e.g. the coughing of the audience, the rustling of programs and creaking of chairs) so the background noise, becomes the music.5 During the same period, technologists and inventors were also experimenting with noise. The major difference was that these technicians were looking at ways to reduce noise in an environment rather than, as some of the examples above show, using noise for artistic purposes.
“In Your Ear: Hearing Art in the 21st Century.” Massmoca.org.
1.2.2 Current Trends In the act of listening, ambient noise often masks the definition of the signal that you are listening for. One active noise cancellation system was developed to combat this problem in the 1950s by Lawrence J. Fogel, and was primarily designed to cancel the noise created in aircraft cockpits. Another system created by Dr. Amar Bose (of the Bose Corporation) was the first set of headphones that utilized similar technology in 1978. This originated from his dissatisfaction with the sound quality produced by headphones in an airplane. These early noise cancelling processes used analog technology. They contained tiny microphones in the earphones that discerned the noise around the listener by inverting the polarity of the undesirable signals in real time. In fig. 3 below, the inverted signals are played into the ear simultaneously in order to cancel out the unwanted ambient noise.
Fig. 3. Noise cancellation illustration. â€œShure SE210A White.â€? Onecall.com. Onecall, 2011. Web. 24 May 2011.
Another noise reducing method that is used in earphones is called Sound Isolation. This method uses a non-active approach to reduce noise. First, a sleeve that is made out of foam and rubber is placed on the ear tips, which allow the earphones to fit more precisely into the ear. The fitting foam and rubber reduces the unwanted noise from reaching the ear canal (see fig. 4).
Fig. 4. Noise isolation illustration. â€œShure SE210A White.â€? Onecall.com. Onecall, 2011. Web. 24 May 2011.
There are some limitations to active noise cancellation. These include the use of an external power source such as batteries, in that they must be regularly replaced to continue working. This also adds extra weight to the earphones. Furthermore, the process of noise reduction can introduce additional noise, such as hiss. They work best with steady, low frequency noises, and they do not attempt to cancel high frequencies. As recording and reproduction techniques evolved in the 1960s, the use of magnetic tape became wide spread within both the consumer and the professional audio environment. One of the major problems that came with the use of magnetic tape was hiss and other noises associated with sound recordings. This greatly diminished the quality of recorded music.
Since music contains a wide variation of dynamics, which can be considered the difference between the loudest sections and the quietest sections, any amount of noise embedded within the recording would be more apparent in the quieter passages. This phenomenon is due to the fact that the ratio between the music signal and the noise decreases, making the audible perception of the noise much greater. To address this problem, noise suppression was developed. Noise suppression is the reduction of apparent noise levels by using dynamic filtering. Whenever a recording is made, undesired sounds such as hiss, hums, pops, and clicks can mask the nuances of recorded sound, annoying and fatiguing the listener. Advances in recording technology increased the problem of noise by producing better microphones and loudspeakers. These tools captured both the signal and the noise elements with higher fidelity, thus creating recordings, which could have, ironically, increased levels of noise. These [noise suppressors] did not alter the loud portions of a recording; instead, they reduced the very high and very low frequencies in the quiet passages in which noise became most audible. However, that removing the high and low frequencies could also affect the desirable portions of the recorded sound. These suppressors could not distinguish desirable from undesirable sounds.6 As recording techniques and technologies improved, another approach to noise suppression was used during the recording process, called sound compression. In 1967, Ray Dolby invented a noise reducer, called Dolby A, which could be used by recording studios to reduce tape signal-to-noise ratios. Later, his Dolby B system, designed for home use, became the industry standard equipment in all types of playback machines: 6
â€œDolby Noise Reduction (Inventions).â€? What-when-how, In Depth Information.
The system operated by carrying out ten-decibel compression during recording and then restoring (noiselessly) the range of the music on playback. This was accomplished by expanding the sound exactly to its original range.7 Dolby A was very expensive, so it was limited to recording studios. In the 1970’s, Dolby invented Dolby B system, the less expensive consumer version. The development of Dolby A and Dolby B noise reduction systems is one of the most significant contributions to the high fidelity recording and reproduction of sound. Therefore, Dolby A quickly became a standard in the recording industry and Dolby B was integrated into almost every high fidelity stereo cassette deck (see fig. 5).
Fig. 5. Dolby noise reduction. Bako, Tomas B. Restoration of Noisy Musical Recordings. Mit.bme.hu. Dolby Labs Incorporated, n.d. Web. 30 May 2011.
These discoveries spurred advances in the field of noise reduction. For example, the German company, Telefunken, and the Japanese company, Sanyo, also implemented their own noise reduction systems. Later, Dolby Laboratories manufactured a better system, Dolby C8. The competition in the area of noise reduction continues, and it will continue as long as changes in recording technology produce new, and more sensitive recording equipment.
“Dolby Noise Reduction (Inventions).” What-when-how, In Depth Information. For additional information see K.F. Louie’s “Dolby Noise Reduction – A Primer.”
1.2.3 Future of Acoustic Noise Cancellation Current research is implementing innovative ways of noise cancellation systems. One of these innovations is a transparent sound-absorbing curtain (see fig. 6). With this invention, there is no longer the need to use heavy curtains in order to avoid noise pollution. Designer Annette Douglas and Weisbrod-Zurrer worked together with Swiss researchers at Empa Institution for Material Sciences and Technology Development. They created lightweight and see-through curtains that are five times more efficient at absorbing sound waves than other curtains available today.9
Fig. 6. Sound absorbing curtains. “Latest Invention: New-Gen Transparent Sound-Absorbing Curtains.” Technology. InfoNIAC.com. InfoNIAC, 05 May 2011. Web. 30 May 2011.
“Latest Invention: New-Gen Transparent Sound-Absorbing Curtains.” Technology. InfoNIAC.com.
Recent research shows that it is possible to generate electricity from noise. A team of students from American University of Sharjah managed to come up with a device that can generate energy from noise, stating, “This device can harness sound power with the help of piezo electric technology that turns noise along with other types of sound energy into electricity.”10 This device works with sound waves that are inaudible to humans (e.g. mechanical energy and ultrasonic waves) and can produce electricity from nearly any type of noise such as traffic or simply footsteps. It can be installed in speed bumps in order to capture the mechanical vibrations and sound waves generated by cars, using the wasted energy to power street lights. It can also be potentially used to power portable gadgets such as cell phones. Another invention combats noise with noise. The constant drone of a computer fan can be annoying. Professor Scott D. Sommerfeldt at Brigham Young University has accepted the challenge to cancel this noise with more noise by producing just the right quantities from small speakers that surround the fan. Prof. Sommerfeldt comments: We make anti-noise. It is the latest example of a technology called active noise reduction, or noise cancellation, well known from its use in headphones designed to block out the low rumble of jet engines.11
“Latest Invention: Device that Generates Electricity from Noise.” InfoNAIC.com. Eisenberg, Anne. "What’s next; To Quiet a Whirring Computer, Fight Noise With Noise."
The sound waves engineered by Sommerfeldt are out of phase with sound waves from the fan and therefore cancel each other out, substantially reducing fan noise. Sommerfeldt's system has four miniature speakers and four small microphones that are set in a circle around the computer fan. The microphones detect the noise of the fan, and with digital signal processing and algorithms, reproduce phase-inverted tones from the speakers. The limitation to this invention is its price. “It's going to be a matter of what the consumer will be willing to pay for a quiet computer.”12 Finally, SONEA (Sonic Energy Absorbing System) is a device that is able to generate energy from any kind of noise that can be used for powering virtually anything. While it is common to use electricity to generate acoustic sound waves, Sonea wanted to do the opposite and produce power from acoustic sound waves. Chris Burns of Yanko Design indicates, “this device is portable, only weights 7 kilograms, and is able to convert 30 watts of power per decibel of sound that it catches” 13 (see fig. 7). During takeoff, an airplane generates a noise around 140dB. By using the Sonea device, it would be possible to produce about 240kW of energy. Considering about 500 airplanes each day, it would be possible to generate about 120MW of energy. In one year, Sonea will be able to generate the amount energy that is equivalent to energy produced by 8000tons of oil.14
Fig. 7. Energy absorbing system. Kim, Jihoon, Boyeon Kim, Myung-Suk Kim, and Da-Woon Chung. “SONEA Converts Sound to Energy.” Yankodesign.com. Yanko Design, 09 Sep 2011. Web. 30 May 2011. 12
Eisenberg, Anne. "What’s next; To Quiet a Whirring Computer, Fight Noise With Noise."
Burns, Chris. “SONEA Converts Sound to Energy.” Burns, Chris. “SONEA Converts Sound to Energy.”
As this current research shows, there are many applications to different types of noise. For the purpose of this research, we shall now examine the relationship between noise and digital signal processing.
Chapter II - Digital Signal Processing 2.1 Types of noise 2.1.1 Background info Noise is a source that is undesirable in a signal. These sources could include: I. Thermal noise on electric conductors II. Shot noise in electric current flows III. Acoustical noise that originates from movement and vibrations Noise can interfere and alter a signal chain. When noise enters a signal chain, it causes some type of distortion (i.e. an undesirable change in a signal). Some of this distortion may fall within the audible range of human hearing. Noise and distortion are the main factors that limit the capacity of any data transmission. This affects the accuracy in signal measurement systems. The modeling and removal of the effects of noise and types of distortion can be achieved by digital signal processing (DSP). The success of a noise reduction method solely depends on its ability to characterize and model the noise. The following chapter explains several different forms of noise, and examines the different types of noise signal processing methods that are used to model them.
2.1.2 Auditory Cognition Several varieties of noise can degrade the quality of a signal. Depending on its source, a noise can be classified into the following: I. Thermal and shot noise II. Channel distortions, echoes & fading III. Processing/Quantization noise IV. Electromagnetic noise V. Electrostatic noise VI. Acoustic noise Thermal noise occurs due to heat in electronic units. This heat is created by the random movement of thermally energized particles in an electrical conductor. It can also be present without any applied voltage. Shot noise is caused by electrons moving and arriving randomly in an electrical unit and is related to current flow. Thermal and shot noises are inaudible until the electronic unit gets hot and malfunctions. For example, the tube inside of a microphone may get hot and distort the frequency response of the microphone capsule. This is the reason why engineers typically place tube microphones upside down (see fig. 8).
Fig. 8. Tube microphone standing upside down. â€?Analog and Digital Recording Studio.â€? Creativecaffeine.com. Creative Caffine, n.d. Web. 4 Apr. 2011.
Channel distortions, echoes & fading occur in telecommunications. For example, as wireless signals reflect off of buildings, they loose their strength as a direct signal, and fade in intensity. Mobile phones are particularly sensitive to this type of noise. While most
Processing/Quantization noise occurs when signals are recorded into a digital Pulse Code Modulation (PCM) file, such as a common AIFF or Wave file. This noise results from the digital to analog processing of signals due to the quantization process and the loss of data during conversion. The resulting noise that is induced is usually unnoticed (see fig. 9).
Fig. 9. Processing noise due to quantization. â€œData Compression Basics â€“ Part 2; Lossy Compression.â€? Soundwave.com. Digital Media Online, 2011. Web. 4 Apr. 2011.
Electromagnetic noise is present at all frequencies. This noise happens particularly at the radio frequency range, where most telecommunications take place. All electric devices, such as radios and televisions, generate electromagnetic noise. In tube televisions, this type of noise is audible as a high-pitched tone. Electrostatic noise is generated by the presence of a voltage with or without current flow. Fluorescent lighting is a common source of this type of noise.
Acoustic noise is common and familiar to everyday environments. It is generated by such sources as moving cars, air conditioners, traffic, walla, wind and rain. These are some of the main factors that prevent a clean production sound recording. These types of noise vary in degree, and the success of removing these undesirable noises depends on the processing methods. Distinguishing a good signal from this type of noise requires finding its characteristics and also modeling the noise process accordingly. All of the different types of noise mentioned above eventually convert into acoustic noise and can be described in three main forms: Impulse, Fixed frequency and Random noise. Impulsive noise consists of short pulse durations. These pulses occur at random amplitude and duration; often sounding like clicks and pops. Fig. 10 shows frequency over time and illustrates the enharmonically placed peaks that represent these clicks.
Fig. 10. FFT of a waveform showing clicks.
Fixed frequency is a narrowband noise, and often occurs as an audible hum in a signal. This hum often occurs between 50Hz to 60Hz â€“ or the first harmonic 100Hz or 120Hz â€“ from the electricity supply. Fig. 11, shows the fist peak at 50Hz and the second peak harmonically at 100Hz.
Fig. 11. FFT of a waveform showing hum.
Random noise can be classified into different categories, and is dependent on the signals frequency spectrum or time characteristics. One form is known as pure white noise. This type is essentially random noise that has a flat power spectrum and contains all frequencies (see fig. 12).
Fig. 12. FFT of a waveform showing random noise.
Pure white noise is a theoretical concept, since it would need to have infinite power to cover an infinite range of frequencies (see fig. 13).
Fig. 13. Constant power spectrum.
On the other hand, band-limited white noise has a flat spectrum and limited bandwidth. Although the concept of white noise provides a reasonably realistic and mathematically convenient approximation to some predominant noise, many other noise processes are non-white. Colored noise is any wideband noise whose spectrum has a non-flat shape such as pink noise or brown noise, also referred to as autoregressive noise. Pink noise (see fig. 14) is commonly used as a reference signal in audio engineering in order to take measurements and calibrate equipment. The power density, compared with white noise, decreases by 3 dB per octave.
Fig. 14. FFT of a waveform showing pink noise.
White noise that passes through a medium is colored by the shape of the frequency response of that medium. For example, most acoustical noise, such as the noise from moving cars, has a colored spectrum. This occurs due to the signals interaction within an acoustical space or environment.
Advancements in technology are eliminating many common problems by modeling noise, using DSP. In recent years, there has been an increase in the development and availability of powerful and affordable digital computers, which aids in the cancelation or reduction of various types of noise. This has been accompanied by the development of advanced DSP algorithms that present a wide variety of applications such as noise reduction, telecommunications, audio signal processing, and pattern recognition. Within the use of DSP, applications of statistical signal processing became a popular research subject. The following chapter examines such issues and its background.
2.2 What is DSP? 2.2.1
Digital signal processing is distinguished from other areas in computer science by the unique type of data it uses; signals. According to Steven W. Smith, PhD: The roots of DSP are in the 1960s and 1970s when digital computers first became available. Computers were expensive during this era, and DSP was limited to only a few critical applications such as radar & sonar for national security, oil exploration, where large amounts of money could be made and also for space exploration.15 Today, DSP relies on microprocessors that are incredibly fast and powerful, and which process data in real time. This real time capability makes DSP perfect for applications such as enhancement of visual images, audio and image recognition, and generation of speech and telecommunication. Within all of these environments, there is little tolerance for delay. DSP utilizes mathematical algorithms to manipulate signals after they have been converted into a digital form.
Smith, Steven W. Digital Signal Processing.
Digital Coding of Audio Signals
In digital processing, analog signals are sampled, and each sample is then converted into a word containing n-number of bits. This digitization process is performed so that the original signal can be recovered from its digital version with minimum loss of information. This conversion consists of sampling and quantization. Sampling is a process that can capture the fastest fluctuations of a signal, when performed with a sufficiently high sample rate. At this sufficient sampling rate (determined by Nyquist’s Theorem), sampling is a relatively loss-less operation. An analog signal can be nearly recovered through interpolation (see fig. 15). Quantization is the process of converting an analog signal into a digital signal, which creates a series of digital values to represent the original analog signal (see fig. 16). The number of bits available (i.e. bit depth) determines the fidelity. Quantization errors can be made negligible by using an appropriately high number of bits.
Fig. 15. Sampling process. “Producing in the Home Studio with Pro Tools.” Music Production School. Berklee College of Music, 2001-2007. Web. 8 Apr. 2011.
Fig. 16. Quantization. “A/D Conversion Quantization Error.” Bb-elec.com. B&B Electronic Mfg, 2011. Web. 8 Apr. 2011.
In digital audio, the memory required to record a signal, the channel bandwidth, and the signal to quantization noise ratio are directly proportional to the number of bits per sample. The objective is to achieve high fidelity with as few bits per sample as possible. There are two main categories of audio coders, model-based and transform-based coders. Model-based coders are used for low-bit-rate speech coding in applications such as cellular phones. Transform-based coders are used in high-quality coding of speech and digital hi-fi audio. After the first step of converting a signal from the analog to digital domain, this signal can now be processed via DSP and potentially improved. These improvements present us with new applications in the reduction of undesirable noise from a signal.
DSP provides the necessary processing power to measure and filter digital representations of continuous analog signals. It is the essential link between digital devices and real-world signals such as light and sound. Today, DSP improves electronic devices such as digital cameras, digital audio and a multitude of new technologies in telecommunications. Because DSP is increasingly faster and more efficient, wireless communications have made tremendous advancements. For example, battery life has become longer, and talk times are now measured in days instead of hours. In audio and music processing, DSP has also introduced advancements in at least three prominent fields: Music reproduction (e.g. CD player), synthetic speech (e.g. computerized voice) and noise cancellation and reduction (e.g. active noise cancelling headphones). For example, in active noise cancellation, the speech signal is observed in an additive random noise, meaning the desirable signal is combined with an undesirable ambient noise such as speech in a noisy acoustic environment (e.g. speech over a moving car or speech over a noisy telephone channel). This information contained within the signal is often contaminated by noise from its surrounding environment. It may be possible to measure and estimate the amplitude of the ambient noise using a directional microphone when using a mobile phone in a moving car, thus subtracting the undesirable signal from the desirable signal. However, in many applications, such as noise reduction during the post-production phase of filmmaking, there is no access to the instantaneous value of the contaminating noise. Only the noise contaminated signal is available. In these cases, the noise cannot likely be cancelled out entirely, but it may be significantly reduced. This is achieved by analyzing the statistical properties of a signal.
2.3 Types of Digital Signal Processing 2.3.1
DSP uses different types of signal processing methods to provide modeling, detection, identification, and utilization of patterns and structures fundamental to the signal. When using the right type of DSP processing, useful information can be obtained about the characteristics, and trajectory of the source signal. Any information from a computer, or any other digital communication device, is in the form of a sequence of binary numbers (i.e. ones and zeros). This information is calculated by a type of DSP and modulated in the appropriate form in order to be transmitted correctly. For example, in the noise reduction of recorded speech, in addition to conveying the spoken word, the acoustic speech signal also has the capacity to convey information regarding the pitch, intonation and stress patterns in pronunciation. In noise reduction, all of these aspects of the signal must be analyzed and calculated in order to preserve the desirable signal and minimize the undesirable signal thereby maximizing the signal-to-noise ratio. This chapter explains the brief history and the development and applications of different types of DSPs that are available for audio processing.
Signal processing methods have evolved in algorithmic complexity in order to achieve the best performance. Depending on the method used, DSP algorithms can be categorized into a combination of any of these three broad categories. These include: I. Model based signal processing II. Transform based signal processing III. Non-linear based signal processing Model-based signal processing uses a parametric model for signal analysis. A parametric model describes the predictable structures and the expected patterns in a signal process. This process aids the forecast of future values of a signal from its past. Some applications of the parametric model are low-bit-rate speech coding, digital video coding, high-resolution spectral analysis and speech recognition. Transform-based signal processing16 includes Fourier transforms and Laplace transforms. The most widely applied signal transform is the Fourier transform, which is a method of transforming continuous signal data into spectral data, where the analyzed signal is expressed in terms of a combination of the sinusoidal waveforms. As Stefan Wรถrner explains,
The history of the Fast Fourier Transform (FFT) is quite interesting. It starts in 1805, when Carl Friedrich Gauss tried to determine the orbit of certain asteroids from sample locations. Thereby he developed the Discrete Fourier Transform, even before Fourier published his results in 1822. To calculate the DFT he invented an algorithm, which is equivalent to the one of Cooley and Tukey.
For a more detailed explanation of FFT, see broadband processors on page 45.
However, Gauss never published his approach or algorithm in his lifetime. It appeared that other methods seemed to be more useful to solve this problem. Probably, that is why nobody realized this manuscript when Gauss' collected works were published in 1866. It took another 160 years until Cooley and Tukey reinvented the FFT.17 The purpose of transform based processing is to convert the signal into a form that provides a more convenient way of interpolation and manipulation. To achieve this, broadband noise reduction processors use an FFT and transform the signal from the time domain to the frequency domain. Then, with the use of an inverse FFT, they are able to resynthesize a cleaner signal. In fig. 17, the input signal is transformed to the frequency domain using Fourier transform and reconstructed using an inverse transform method.
Fig. 17. Illustration of a transform-based coder. Vaseghi, Saeed V. Advanced Digital Signal Processing and Noise Reduction. West Sussex: John Wiley and Sons, 2006. Print.
There are two main advantages of coding a signal in the frequency domain. Firstly, the frequency domain structure of the FFT allows the user to pinpoint the offending frequencies with relative precision. Secondly, the frequency bins can be adjusted completely independently of one another, unlike traditional analog filtering. 17
Wรถrner, Stefan. Fast Fourier Transform at the Numerical Analysis Seminar.
Fourier transforms are being used in common music coders, within noise reduction tools, and for signal pattern recognition. Non-linear signal processing or Bayesian signal processing is a theory used to generalize the estimation of random processes such as those that occur in white noise. The fluctuations of this random signal cannot be modeled by a predictive equation. Instead, they can be defined using statistical average of a signal's values. Thomas Bayes (1702-1761) is the mathematician who proved Bayes’ theorem. However, it was Pierre-Simon Laplace (1749–1827) who introduced a general version of the theorem, Laplace transform, and used it to approach problems in mechanics, medical statistics and reliability.18 The Laplace transform recalculates the equation each time there is a new set of data and is able to calculate a better probable hypothesis from less valid ones. The ideas of Laplace were further developed in two different directions, objective and subjective Bayesian practice. Objectivists depend solely on the model assumed and the data analyzed. No subjective decisions are involved. On the other hand, subjectivist statisticians deny the possibility of a completely objective analysis. Not only is Bayes’ theorem used in estimation of random noise, but it has also revolutionized robotics technologies. One of the latest inventions that utilize Bayes’ statistical calculation is Google’s driverless car (see fig. 18).19
Candy, James V. Bayesian Signal Processing: Classical, Modern and Particle Filtering Methods. McGrayne, Sharon Bertsch. “Why Bayes Rules: The History of a Formula That Drives Modern Life.”
Fig. 18. Google driverless car. Rickman, Ollie. â€œWhen could consumers buy a Google driverless car? And why would they?â€? Tech.fortune.cnn.com. Cable News Network; A Time Warner Company, 12 Oct. 2010. Web. 8 Apr. 2011.
This invention illustrates that there are new and even better ways to apply a 250-year-old theorem.
With advancements in technology and improvements in mathematical algorithms, we can characterize desirable signals from undesirable signals much more accurately. One of the primary uses of DSP can now be used to reduce interference, noise, and other undesirable components in a noisy production sound recording. By using statistics and probability, we allow these undesirable features to be measured and classified. We can then identify the strategies to remove the offending noise within a signal. The following chapter introduces the most important concepts in statistics and probability with an emphasis on how they apply to the noise reduction application, CineClean Pro.
Chapter III â€“ Background of CineClean Pro 3.1 Introduction Many modern noise reduction plug-ins work by analyzing the differing statistical properties in a signal and the noise components therein. The identification of the sources of noise is the single most important aspect of this process. Different types of noise require different methods to remove them. The development of DSP techniques for noise reduction offers several applications, and the research on this subject is now popular in the field of engineering. In order to characterize noisy signals and the processes that generate them, DSP uses statistical and probabilistic methods. This chapter introduces the most important concepts in statistics and probability and how they relate to noise.
A probability function takes a variable with some value as an input; while the output gives us the probability of the variable having that value. Within this function, every value must be between 0 and 1 and the probabilities of all the values must add up to 1. For example, if there are two sources for a signal, Noisy (N) and Clean (C), variable x has two values: N and C. If the probability that x=N is 0.5, then the probability that x=C is also 0.5. This will be written as: P (N) = 0.5 P (C) = 0.5 When predicting noise values, random chance will make the probability of noisy signals over clean signals slightly different each time as the experiment is repeated. This metairregularity is often referred to as statistical variation, fluctuation, or noise. In a probability function, a signal is a description of how one parameter is related to another. In an analog signal, voltage varies with time. This signal is referred to as continuous since both parameters, in this case, voltage and time, have a constant and indefinitely precise range of values. When an analog signal gets converted to a digital signal (e.g. PCM digital audio), it becomes discrete. This happens due to the conversion process (i.e. each of the two parameters become quantized). When a DSP analyzes a signal, the vertical axis (y-axis) may represent voltage, sound pressure or amplitude while the horizontal axis (x-axis) shows other parameters of the signal such as time. Time is the most common parameter implemented on the horizontal axis. If a signal uses time as its horizontal axis then that signal is said to be in the time domain.
If a signal implements frequency as the variable on the horizontal axis, then the signal is referred to as the frequency domain. Conversion of analog signals to digital PCM format allows us to easily calculate the following: I. Mean II. Standard deviation III. Statistical signal-to-noise ratio IV. Root mean square Mean is found when all of the samples are added together and divided by the total number of samples. It is indicated by lower case Âľ (see fig. 19).
Fig. 19. Mathematical equation for calculating the mean.
If the signal is a simple sine or a square wave, it can be described by the mean of its peak-to-peak amplitude.
Standard deviation of random noise signals do not show defined peak-to-peak value. In this case, calculating the standard deviation of a signalâ€™s data is used to characterize the signal. Standard deviation is indicated by a lower case Greek sigma (Ďƒ) (see fig. 20).
Fig. 20. Calculation of the standard deviation of a signal.
The standard deviation measures how far a signal varies from its mean. The mean describes what is being measured, while the standard deviation represents noise and other interferences. Statistical signal-to-noise ratios can be calculated by dividing the mean by the standard deviation. The standard deviation here is important only when compared to the mean. This comparison is called signal-to-noise ratio (SNR)20. Root mean square (RMS) is calculated by three easy steps: squaring all the values, averaging these values over a complete cycle, and finding the square root of this average.21 RMS measures both the AC and DC components, while the standard deviation measures only the AC portion of a signal. If a signal has no DC component, its RMS value is identical to its standard deviation. If a signal has no AC component then its RMS value is equal to its peak value (e.g. square wave).
This SNR is different than what recording engineers typically consider SNR, in that it is a statistical measurement, rather than an acoustical measurement. 21 Lewis, John. "AC, DC, and Electrical Signals."
Statistical Noise Modeling
Statistics is the science of interpreting numerical data. This interpretation can then be used to make decisions. In comparison, probability is used in DSP to understand the processes that generate signals. Although they are closely related, the distinction between interpreting the numerical data and understanding the underlying processes is fundamental to many DSP techniques, particularly in noise modeling. Important theories used for statistical noise modeling include: I. II. III.
Gaussian process Hidden Markov models Bayesian estimation theory
Gaussian process is used and applied widely, and can model the distribution of various types of random processes including noise. The results of the Gaussian estimation method are usually linear due to the probability that the signal has a defined value. A set of assumptions aid in defining noise as a stationary additive white Gaussian noise (AWGN). The first assumption is that noise is additive, meaning that the received signal is equal to the transmitted signal with some additional noise. Second, is that the noise is white, meaning that the power spectral density is flat. In other words, the signal contains equal power within a fixed bandwidth at any center frequency (see fig. 21).
Fig 21. White noise with a flat frequency spectrum. â€œNoise Modeling.â€? Deonarain, Danesh, Georgina Mang, Teresa Misiti, and Carolyn Ehrenberger. controls.engin.umich.edu. University of Michigan, 16 Oct. 2010. Web. 20 Apr. 2011.
Although for some situations these assumptions are valid and aid in the solution of mathematical formulas, in practice (e.g. location sound recordings), noise is often time varying. Not only does noise affect other variables, but it also depends on other variables, which makes it non-linear. This is particularly true for impulsive and acoustic noise, which are both non-stationary (i.e. independent of any predictive or constant value) and non-Gaussian, and therefore cannot be modeled using the AWGN process. Nonstationary and non-Gaussian noise can be modeled by a Hidden Markov model. Hidden Markov models (HMM) are used for the statistical modeling of non-stationary signals such as speech signals and time-varying noise. The Markov process, developed by Anderi Markov, is a process whose state or value at any time, t, depends on its previous state and values at time tâˆ’1, and is independent of the history of the process before tâˆ’1.22 One of the most popular forms of the Hidden Markov model is the left to right model. This model is useful to analyze speech and music signals in the time domain from left to right. An HMM is essentially a Bayesian finite state process. While the Bayesian model calculates only with initial and prior knowledge of the signal, an HMM can model the time variations of a random noise by using a probability density function for modeling within each signal state. This type of modeling of statistical variations can be useful for acoustical understandings (e.g. the removal of noise within human speech).
Vaseghi, Saeed V. Multimedia Signal Processing: Theory and Application in Speech, Music and Communications.
Bayesian estimation theory is used to formulate statistical problems. It analyzes disturbances contained in a signal with the prior knowledge of the probability distribution of the noisy signals, gathering conclusions from evidence. In practice, this model of analysis works by comparing between the stable and unstable components (i.e. desirable and undesirable components) of a signal over time. This allows the model to understand the probability distribution of unstable components (e.g. noise), therefore allowing the process of identification and removal. One popular method is called Bayesian Spectral Amplitude Estimation. This method is often implemented in speech restoration and recognition. In order to identify noise, it uses a spectral subtraction method by utilizing the probability density function of that signal. This method is used to cancel additive noise by subtracting the average noise amplitude from the signal. One problem with this method is that non-linear processing of the signal may fall below a set threshold level. This may introduce a different type of noise (e.g. processing noise) back into the original signal.
Engineering and scientific applications such as noise reduction plug-ins require many statistical calculations. Formulas and calculations required to analyze random disturbances like noise can be generated using Matlab software (see fig. 22), a technical computing language for algorithm development. Mathematician and creator of MATLAB, Cleve Moler, decided to create a software other than the FORTRAN (i.e. the dominant computer language for scientific community), which was easier to input and output data for numerical computing. MATLAB has quickly become the standard tool in both professional and academic circles. It is widely used within engineering, biology, telecommunications, statistics, and signal-image processing.
Fig. 22. Illustration of Matlab. “MATLAB – The Language of Technical Computing.” Mathworks.com. The Mathworks Inc., 1994-2011. Web. 20 Apr. 2011.
There are endless variations of calculations available in Matlab, which allow engineers to calculate and model statistical noise. Today, in the audio industry, Matlab is where technology meets creativity. Matlab and other companies try to create new or improve on existing algorithms in order to remain competitive in today's new media market. With today's technology, new algorithms and new ways to cancel noise are being shared exponentially, and are stored in libraries such as Linpack and Lapack. There are now more people than ever collaborating to win the war on noise.
Current Noise Reduction Plug-ins 3.2.1
Over the last twenty years, DSP and noise reduction software has developed immensely. Noise reduction plug-ins are now so popular that the 21st century audio engineers/artists are familiar with more than one type. Until the late 1990s, artists were hesitant to make the switch from analog to digital. However, they quickly discovered the benefits of being able to work on multiple digital audio workstations (DAWs) and a selection of noise reduction plug-ins. In addition, with the help of onboard DSP chips (e.g. Universal Audio), engineers are now able to emulate popular analog audio hardware such as dynamic filters, equalizers and many more audio restoration tools. Today, a modern audio post-production facility typically consists of a Mac or PC computer that has multiple DAWs with sufficient DSP power. DAW software makes it possible to seamlessly and non-destructively use a noise reduction plug-in to edit and clean digitized audio files. Current noise reduction plug-ins contribute to the film industry by providing sound engineers with the ability to use a selection of tools, thus providing the option of limitless “undo” and “redo” on any given instance of an audio file. The following chapter introduces the most important concepts and tools that are available in today's professional audio plug-ins with an emphasis on how they apply to location sound noise reduction.
During the post-production process, several tools can be used as a first line of defense to clean a noisy location sound recording. These tools include: I. Parametric equalizer II. Gate III. Dynamic filters IV. Interpolation processors V. Broadband processors VI. Spectrum canceling Parametric equalizer works within the frequency domain. It can be employed to reduce the hum within an audio signal by simply filtering out the 60Hz and its harmonics from the frequency spectrum. With a low-pass filter, parametric equalizers can reduce excess hiss that may build on higher frequency ranges. With a high-pass filter cut off point around 120HZ, microphonics (i.e. mechanical vibrations inducted into the microphone), wind, or any low frequency rumble may be removed. Gates are a type of expander and they work dynamically. Expanders can make a quiet signal even quieter by reducing the portion of the signal that falls below the set threshold level. If there is a certain amount of signal above the set threshold point, the gate opens and allows the signal to pass. If not, it closes and cancels the noisy background from being heard.
Dynamic filters work by moving the cut-off frequency dynamically according to the signal content. They can attenuate random noise by ‘ducking’ (i.e. masking) the high frequency content. They are often used in movie dialogue to reduce noise. One disadvantage of this technique is that the filter tends to round off transients thus obscuring the original signal. Interpolation processors can estimate lost samples of a signal by analyzing the average number of known samples next to the unknown samples. These processors first identify the sounds that have – unnaturally short attack and decay times (i.e. clicks and pops) then they remove that sample, scan left and right of that event and interpolate to the next audio sample. Interpolators are mostly used in various forms of communications signal processing and also decision-making systems. They are also a common technique to be used in movie dialogue, especially for lavaliere microphone recordings that often suffer from the noise created from actor’s clothes. But in doing so, they have the potential to reduce transient fidelity. Broadband processors are essentially multiband expanders and use Fast Fourier Transform Filtering, FFT, for noise reduction. FFT is a more reliable method of removing all noise artifacts in one process. When an audio signal is analyzed with FFT, the signal is moved from the time domain to the frequency domain. In the frequency domain, this FFT of the signal is then divided into narrow-frequency bins. This enables the elimination of the noisy bin by an inverse FFT process. Broadband processors often allow a creation of noise signature by selecting and analyzing an example of the noise within the signal (i.e. a section of a location sound recording that contains only the noise).
This signature serves as a blueprint of noise and makes it easier to fine tune adjustments for maximum noise reduction and minimum signal loss. In contrast, they might introduce undesirable artifacts that may result from the inverse FFT processing and they also have long latency due to intensive processing. Spectrum canceling or editing can isolate and eliminate unwanted signals visually, even if they are masked by complex noise or other harmonically multiplying frequencies. Again, FFTâ€™s are used. The spectral domain visualizes the signal, allowing the use of both eyes and ears to identify and remove undesirable artifacts. Editing the undesirable components with a visual representation helps to preserve the harmonic components (e.g. human speech), nonlinear pitch changes (e.g. saxophone vibrato), and transients. This allows the re-synthesizing of the signal with an inverse FFT and avoids over processing of the signal.
With all of the noise reduction methods, applications, and processes available, artists are now better equipped to control noise on production sound. With the help of many noise reduction plug-ins, new algorithms, and technological developments, a noisy recording can now be cleaned at a post-production facility, or even at a home studio. Still, this process is very detailed, and can take a significant amount of time and resources in an industry where both are scarce. Even more troublesome is the fact that audio postproduction typically occurs in the final stages of post-production before the media is released. All of these factors effect the quality of new media art, especially independent film and video. The following chapter explains in detail the proposal of CineClean Pro, a new noise reduction application, which will increase efficiency and quality in postproduction workflow.
Chapter IV â€“ Development of CineClean Pro 4.1 Business Model 4.1.1
With the advent of new technologies, and an increase in video content being created by filmmakers, it is important to the development of visual media to have the tools to yield high quality audio. Currently almost all content is produced digitally. With the prices of digital cameras decreasing and the quality of the image rapidly increasing, a large amount of new media is now created, as evidenced by the vast amount of videos uploaded to social networking sites every day. Current consumer-level audio recording hardware is incapable of recording clean sound, often producing thermal noise because of poor preamp designs, for example. Even high-end professional equipment is prone to acoustic noise, be it from environmental factors or poor recording technique. There are many other factors that result in poor quality audio. Low-budget and no-budget productions offer limited resources for high fidelity location recordings and limited time for audio post-production. It is therefore imperative to have a fast processing noise reduction application that is versatile and powerful, while at the same time intuitive for the average editor.
CineClean Pro will provide a comfortable environment where editors can be more efficient and more effective throughout the noise reduction process. From directly within the Final Cut Pro application, editors will be able to simply drag and drop noisy audio files from their timeline into CineClean Pro for analysis. Once in CineClean Pro, the editor will be able to select from a series of icons that represent the environmental conditions under which the audio was recorded. With the sensitivity control, the editor will be able to visually apply the correct noise reduction techniques and necessary algorithms to clean the undesired artifacts. CineClean Pro will duplicate referenced media files, providing non-destructive processing. Processing will affect the entire duplicate file, thus giving the user the opportunity to pull out handles once the processed file is brought back into FCPâ€™s timeline. Since noise will no longer be a distraction, the editor will be able to shape the story using not only his eyes, but also his ears.
The following is the SWOT (Strengths, Weaknesses, Opportunities, and Threats) analysis of CineClean Pro in order to understand the market demand better. Strengths – Graphical images will allow the user to relate to different types of noise disturbances. There will be a single fader to control the sensitivity level of noise reduction. The program will offer direct import and export of files from within the Final Cut Pro timeline or browser. Presets will be designed to eliminate most common types of unwanted noise including traffic, wind, hum, hiss, crackle, and pop. Additionally, the graphical user interface (GUI) will be based on Final Cut Pro’s design and navigation. Weaknesses – CineClean Pro will not offer seamless compatibility with other video editing programs, such as the industry standard, AVID Media Composer. Compatibility with future Final Cut Pro software updates cannot be determined; the updates will be available only if there is demand. The GUI will have fairly basic control parameters, which may reduce the ability to manipulate noise reduction at a detailed level due to the limited number of algorithms, and therefore presets, that will be included in the program. Opportunity – CineClean Pro will be free to use, and will be available immediately via online download. With the open source structure, users will be able to upload their own presets, automatically creating a collaborative environment. Connecting a broader media community by sharing information will maximize efficiency of workflow. This will offer the opportunity to expand into additional tools for the post-production process.
Threats â€“ Recent negative reviews of Final Cut Pro X may result in the discontinuous use of the Final Cut Pro editing software for post-production, rendering CineClean Pro obsolete under this initial proposal. CineClean Pro will be operational only in the Final Cut Pro environment, but existing noise reduction applications such as Waves X-Noise or iZotope RX, can be applied to FCP or any other DAW, as they are multi-platform. Additionally, advancements in field recording hardware may reduce or eliminate the need for post-production noise reduction.
CineClean Pro will have a familiar GUI design for current Final Cut Pro editors, essential for the seamless integration with FCPâ€™s post-production workflow (see fig. 23). Precious post-production time will therefore not be spent learning CineClean Pro, but using it. The graphical representation of the audio waveform and navigation will be the same as that of Final Cut Proâ€™s, which will make the user interface easy for editors to manipulate. The interface will be intuitive by having one panel on the left named dirty bin and another panel on the right named clean bin. The user will be able to drag and drop audio files of various formats, then process the file, and export the file, all from within the same window. There will be two faders built onto the audio level meters to control and indicate the amplitude levels and clipping on both the input and output of the signal. Additionally, there will be six types of built in processes in the main window, easily accessible, each with visual representations of common location noise problems.
Fig. 23. Illustration of CineClean Pro. DeVault, Talia. CineClean Pro. 2011. Cinecleanpro.com. Web. 13 Aug. 2011.
From directly within Final Cut Pro, the user will be able to drag and drop audio files into the clean bin. The new audio file will be a duplicate of the parent file, and this nondestructive process will provide the user with “a road map back home”.23 The graphical representation of the audio waveform will be displayed by clicking on the carrot directly below the dirty bin. The user will be able to preview the file, zoom, and navigate to problematic areas as well as use the loop feature to preview a selection. If the user makes a selection of the waveform, only the selection will be processed. However, to process the entire file, no selection is necessary. By using the six preset selections, the user may apply one or many algorithms for noise reduction. By control clicking the ‘hiss’ preset, the user can enable the ‘learn’ option in order to create a blueprint of the noise. The input fader will allow the user to set the level of the incoming signal. The sensitivity fader will enable the user to increase or decrease the strength of noise reduction within set values. Once the desired effect is reached and the output fader is set to the desired level for export, the user will click on the process button, which will provide a processed version of the audio file in the clean bin. By holding the control button and clicking on the audio file in the clean bin or dirty bin, the user may easily audition and compare the unprocessed and processed files respectively. Finally, the user will be able to then drag and drop the processed file directly onto Final Cut Pro’s timeline.
Makes it possible to reference or replace the original file. Ryan Kleeman, Instructor at The Art Institute of California at San Francisco
CineClean Pro offers six presets; traffic, wind, hum, hiss, crackle, and pop. The traffic preset acts as a noise suppressor, and will reduce traffic noise from dialog through the use of dynamic filters. The filter’s frequency bands will target low frequencies between 20Hz to 300Hz with additional attenuation in the low-mid frequencies between 300Hz to 1000Hz. This will identify and reduce traffic noise with minimal artifacts on the dialog. The sensitivity fader will control the threshold level (i.e. the amount of signal to be processed). The wind preset will use a parametric equalizer, set as a high-pass filter. The sensitivity fader will control the frequency cut-off point of the filter, and will be limited between 20Hz to 200Hz in order to avoid excessive removal of low-frequency content, which can ‘thin’ human speech. The hum preset will use a linear phase notch equalizer, and will automatically identify the undesirable fixed frequency of the hum by analyzing the 50Hz to 60Hz frequency bins via an FFT. The sensitivity fader will control the amount of noise reduction that will be applied to the signal. This preset will be great to use for the precise removal of hum’s fundamental frequency and its harmonics. The hiss preset will use a broadband noise reduction processor. With the use of the learn command, it will analyze a section of the location sound recording that contains only the noise by creating an FFT profile of the noise shape. This FFT profile will create frequency bins, which will be attenuated selectively according to the threshold level. This will provide clarity and more control over components of noise. The sensitivity fader will control the threshold level. The crackle preset will use interpolation process in order to automatically identify and remove randomly distributed continuous clicks within a signal. The sensitivity fader will control the threshold level, which will set the amplitude for crackles that are targeted for removal.
As threshold values increase, crackles with smaller amplitude will be processed. However, higher threshold values will remove excessive transients, which might introduce smoothing of high frequency content. This preset is ideal for noise artifacts that are often produced by actors’ clothing rustle, which can distort the microphone’s capsule. The pop preset is designed to reduce low frequency pulses that often occur when talking closely into a microphone (i.e. distorting the microphone capsule) and from combining sections of wave files together. This preset’s algorithm will target low frequency sound waves, which have unusually short attack and decay times in the time domain, and will use a combination of non-parametric linear filters.24 These filters will only pass frequencies below a certain value that have slow-moving components in a signal’s time domain. The average values that result from the filters will be used to identify and remove strong impulsive errors without audible artifacts. The sensitivity fader will control the amount of attenuation being applied by the filter.
A non-parametric filter can be designed to observe a specific noise disturbance thus does not require any statistical data from the noisy signal. Taking average values from a combination of non-parametric linear filters minimizes the difference between the actual and predicted outcome.
Both the design and application of CineClean Pro will allow a multitude of editors using Final Cut Pro to enhance their production audio. As a result, editors will be able to compete in todayâ€™s rapid marketplace and rigorous deadlines. And, since the user interface and the design will be so familiar to the target market, integrating this application into the everyday post-production workflow may become seamless. A seamless workflow not only saves time, but also can prevent stress, and even increase the number of projects or clients.
Chapter V - Conclusion Noise is everywhere. We live in and around environments that are constantly filled with undesirable energy. For most, this goes unnoticed, since we primarily rely on our visual sensory cues rather than auditory cues when we analyze and adjust to new environments. On a film set, when the film crew arrives on location, the multitude of departments and people dedicated to the visual aesthetic of the film begin setup and overwhelm the space. There are many discussions about camera placement, angle, movement, lighting, art direction, and even make-up, but not often sound. Locations are not typically chosen for their unique soundscape, yet they are not discarded because of excessive noise. Even more troublesome, filming continues regardless of traffic or wind, or because of noise from the generators powering the set. Should there be an airplane flying overhead in a romantic outdoor park scene, the director still chooses the scene for the final cut. Unfortunately, later in the cutting room, the director realizes that the scene is unusable as is, due to the excessive noise. Additionally, the budget, schedule, unions, and permits limit resources and do not encourage an optimal recording environment. The film reaches the final stages of post-production with noisy or even indiscernible dialog. It is only after the film is cut that the undesirable sound is considered. All of these factors prevent high fidelity sound, especially in independent film and video. Even if a Director or film relies heavily on sound throughout all stages of the film, the environment continues to pose a threat. Again, noise is everywhere. Regardless of the quality in sound recording equipment, noise reduction will be necessary to remove undesirable signals from location sound recordings.
With improvements in mathematical algorithms, characterization of desirable signals from undesirable signals is now more accurate, and, we may eliminate these undesirable signals and noisy backgrounds at an earlier stage of post-production. With the help of CineClean Pro, artists will be better equipped to control noise on production sound. Editors will now be able to enhance their production audio from within Final Cut Pro, and will be able to shape the story while listening to a clean dialog. CineClean Pro will offer a creative solution to reduce noise in location sound recordings, thus creating a more enjoyable presentation and experience for the audience. “We do not see and hear a film, we hear/see it.” – Walter Murch.25
Murch, Walter. “Stretching Sound to Help the Mind See.”
Works Cited Bells, Mary. “Mary Walton”. About: Inventors. About.com., 2011. Web. 04 May 2011. Burns, Chris. “SONEA Converts Sound to Energy.” Yanko Design, 09 Sep. 2009. Web. 30 May 2011. Candy, James V. Bayesian Signal Processing: Classical, Modern and Particle Filtering Methods. Online: Wiley-Interscience, 06 Apr. 2009. “Dolby Noise Reduction (Inventions).” What-when-how, In Depth Information. TheCrankshaft Publishing, 2011. Web. 30 May 2011. Eisenberg, Anne. "What’s next; To Quiet a Whirring Computer, Fight Noise With Noise." The New York Times [Online] 27 May 2004: Web. 30 May 2011. “In Your Ear: Hearing Art in the 21st Century.” Massmoca.org. Mass MoCA, 2011. Web. 24 May 2011. Kosko, Bart. Noise. New York: Penguin, 2006. Print. “Latest Invention: Device that Generates Electricity from Noise.” InfoNAIC.com. InfoNAIC, 08 Apr. 2011. Web. 30 May 2011. “Latest Invention: New-Gen Transparent Sound-Absorbing Curtains.” Technology. InfoNIAC.com. InfoNIAC, 05 May 2011. Web. 30 May 2011. Lewis, John. "AC, DC, and Electrical Signals." Kpsec.freeuk.com. The Electronics Club, 2011. Web. 24 Feb. 2011. “Luigi russolo.” Wikipedia 02 Sep. 2011. Wikipedia. Web. 24 May 2011. McGrayne, Sharon Bertsch. “Why Bayes Rules: The History of a Formula That Drives Modern Life.” Scientific American: Scientific American Magazine (15 May 2011): n. pag. Web. 08 Apr. 2011.
McLoughlin, Ian. Applied Speech and Audio Processing. Cambridge: Cambridge University Press, 2009. Print. Murch, Walter. “Stretching Sound to Help the Mind See.” Filmsound.org. n.p., n.d. Web. 06 Apr. 2011. Peterkin, Tom. “Noise Pollution Map Warns of Health Risks.” The Telegraph. 16 May 2008: 400 BST. Web. 29 May 2011. “Producing in the Home Studio with Pro Tools.” Music Production School. Berklee College of Music, 2001-2007. Web. Smith, Steven W. Digital Signal Processing. Burlington: Elsevier Science, 2003. Print. Vaseghi, Saeed V. Multimedia Signal Processing: Theory and Application in Speech, Music and Communications. New Jersey: John Wiley and Sons, 2007. Print. ---. Advanced Digital Signal Processing and Noise Reduction. West Sussex: John Wiley and Sons, 2006. Print. Wörner, Stefan, Bachelor. Fast Fourier Transform at the Numerical Analysis Seminar. Swiss Federal Institute of Technology. Zurich. 2008. Lecture.
References Bitzer, Joerg, and Matthias Brandt. “Speech Enhancement by Adaptive Noise Cancellation: Problems, Algorithms, and Limits.” 39th International Conference: Audio Forensics: Practices and Challenges (June 2010) : 3-3. Web. 02 Apr. 2011. Dolby, Ray. An Audio Noise Reduction System. AES E-Library: Audio Forensics: 33 (October 1967) : 543. Web. 06 Apr. 2011. Dutoit, Thierry, and Ferran Marques. Applied Signal Processing. New York: Springer Science, 2009. Print. Eagle, Dougles Spotted. “The Art of Noise Reduction.” Digital Media Online. 2010. Web. 05 May 2011. Ives, Fred H. “A Noise-Reduction System: Dynamic Spectral Filtering.” AES E-Library: JAES: 20.7 (September 1972) : 558-561. Web. 08 Apr. 2011. Kallianpur, G., and R. L. Karandikar. “White Noise calculus and Nonlinear Filtering Theory.” The Annals of Probability. Vol. 13, No. 4 (Nov., 1985): 1033-1107. JSTOR. 18 Feb. 2011. Kroschel, K., M. Ihle, M. Kuropatwinski, and Andrzej Czyzewski. “Adaptive Noise Cancellation of Speech Signals in a Noise Reduction System Based on a Mircophone Array.” AES ELibrary: 102 (March 1997): 4450. Web. 04 Apr. 2011. Louie, K.F. “Dolby Noise Reduction – A Primer.” Pw1.netcom.com (1995). Web. 20 Apr. 2011. Mantysalo, S., and J. Vuori. “Effects of Impulse Noise and Continuous Steady State Noise on Hearing.” British Journal of Industrial Medicine. Vol. 41, No. 1 (Feb., 1984): 122-132. JSTOR. 18 Feb. 2011. McMahon, David. Matlab DeMystified. New York: McGraw Hill, 2007. Print.
Noor, Ali O. Abid, and Samad Salina Abdul and Hussain Aini. Development of a Background Noise Cancellation System Using Efficient Oversampled DFT Filter Banks. Australian Journal of Basic and Applied Sciences 3.2 (2009): 1185-1197. Purcell, John. Dialog Editing. Oxford: Elsevier Inc., 2007. Print. Roads, Curtis. The Computer Music Tutorial. The MIT Press, 27 Feb. 1996. Print. Shenoi, B. A. Introduction to Digital Signal Processing And Filter Design. New Jersey: Wiley, 2006. Print. “Using the FFT.” Cycling 74.com. Cycling 74, 2011. Web. 23 Feb. 2011. “What is DSP?’ Texas Instruments Inc., 2001. Web. 05 May 2011.