Acoustics for Electronic Music Venues. Bachelor Project. Tsvetan Nastev

Transcription

1 Acoustics for Electronic Music Venues Bachelor Project Tsvetan Nastev December 30, 2013

2 Abstract The acoustics of a venue as well as the quality of the PA sound system and medium are responsible for delivering quality sonic experience to the audience. In order for good sound to be achieved, each part of the chain has to be at its best. Starting from the music medium, through the signal path, then translated into acoustical energy, the sound is being altered before it reaches the human ear. It is important to take advantage of the possibilities of today s systems to record and reproduce sounds with a dynamic range higher than ever before and keep it distortion less. In room acoustics, sound decay is of great importance. Electronic music has a heavy low end content. Low frequencies are favored by listeners and essential for a whole and natural experience. The unnatural square layout of most of todays rooms as well as the materials used in construction create acoustical problems. Standing waves cause lack of definition in the low end, as high end reverberation is being damped by conventional materials. Sound environments that are closer to natural are the more preferred by people, they enhance the spatial translation of the PA system.

3 Contents 1 Introduction 1 2 Theory Odeon Reverberation Parameters Auralisation Binaural Recording Binaural Room Impulse Response Problem Description Signal Path Perception Reverberation Simulation Music Samples Default Room Fixed Low End Reverberation Hearing Curve Room Inverse Hearing Curve Room Reflective Side Wall Room Reflective Front and Back Wall Room Concrete Room

4 5 Processing Deconvolution Convolution Measurements Question Question Question Question Testing Program Statistics Question Question Question Question Discussion 53 8 Conclusions 56 9 Future Improvements 58

5 Chapter 1 Introduction The aim of this project is to produce a guide for improvement of the perceived sound quality in music venues. The guide includes basic tips on how the different links in the audio chain affect the sound quality as well as how to prevent unnecessary distortion. The guide will also introduce brief theory on human sound perception and tips on how to improve the existing acoustical environment. This is realized by investigating on how original audio signals are being altered through the course of the signal path - from the medium, to the customers ears. A research in the field of electronics is performed in order to find the bottlenecks of today s digital PA systems and a statistical field research is conducted in order to find how sound reverberation is perceived in different acoustical environments. Wrapping the two concepts together gives a good understanding on the basics of sound and acoustics for electronic music venues. 1

6 Chapter 2 Theory 2.1 Odeon The simulation of different acoustic environments is based on convolving the expected binaural impulse response of the given room with the desired music piece. To achieve this, a software program Odeon is used to create the binaural impulse response files. The Odeon software is able to import 3d models and then assign materials to all the surfaces. The emphasis is on how reverberation time and frequency response affect the subjective listening experience of people. 2.2 Reverberation Parameters One of the basic measurements of the reverberation time of a room is the T30 which measures the time it takes for the sound pressure level to drop by 30 db. Measuring the T30 at different frequency bands gives a clear understanding of which parts of the audible spectrum are being amplified or damped by the room. Varying the T30 at different bands during the statistical experiment can supply information on what kind reverberation is favored by the human hearing system. Another reverberation measurement is the EDT or Early Decay Time which is used to determine the first reflections caused by the rooms surfaces. It is obtained from the initial 10 db drop of the sound decay. Clarity C80 is yet another parameter which is a measurement of the difference in energy contribution between the first 80 milliseconds and and the time slot from 80 until the complete decay. 2

7 2.3 Auralisation Auralisation is the process of rendering audible, by physical or mathematical modelling, the sound field of a source in a space, in such a way as to simulate the binaural listening experience at a given position in the modelled space. When used in the computer simulation program ODEON, auralisation is considered as the art of creating digital simulations of binaural recordings in rooms which may not be build yet. The aim is to provide the same threedimensional listening experience to the listener as would be achieved in the real room at the given receiver position with the simulated source positions and signals. Figure 2.1: Auralisation Reflectogram 2.4 Binaural Recording Humans usually listens using two ears in most cases. This allows us to perceive sound as a 3D phenomenon. To create a binaural recording, it is not enough to create a two-channel recording (stereo), also the coloration created by diffraction from the human body has to be included. This is usually done by using a dummy head with a microphone mounted at the entrance of each ear canal - this recording may be recorded using an ordinary stereo recorder - but is now refereed to as binaural. Binaural recordings are usually played back through headphones to avoid coloration from the room in which it is played as well as avoiding diffraction from the human body to be included twice (at the recording and at the playback). If one has measured or indeed simulated the binaural room impulse responses in a room, it is possible to simulate a binaural recording. 3

8 Figure 2.2: Binaural Room Impulse Response 2.5 Binaural Room Impulse Response The binaural room impulse response is the key to binaural room acoustic auralisation. The BRIR is a set of impulse responses detected at the left and right entrance of the ear canals of a dummy head (or indeed at blocked entrenches of the ear channels of a (living) person residing in a room, when a sound source (or some sound sources) has emitted an impulse. The BRIR should include all the (necessary) information on receiver position and orientations, source(s) position(s) and orientations, room geometry, surface materials and the listeners geometry. Convolving the left channel of the BRIR and the right channel of the BRIR with a mono signal, a binaural signal is created, which when presented to the listener over headphones gives the impression of the three dimensional acoustics at a particular position in the room. It is also possible to simulate the recording of the binaural room impulse responses, which is what the coomputer simulaiton program ODEON does. 4

9 Chapter 3 Problem Description 3.1 Signal Path Before the signal becomes sound, it is being converted back and forth between various domains. Depending on the quality of the converters that are used, the signal can either degrade more or degrade less. A typical DJ setup would look something like this: CD Record A/D Convertor Cartridge Mixer DSP Amplifier Speaker Figure 3.1: Signal Chain The first block represents the medium. It can be either a record, or a magnetic tape, a compact disc or another digital device. In order to be further processed the medium has to be read. Be that a cartridge that will transfer the mechanical movement to electrical signal, or a tape head that will pick up the magnetic track or a laser that will read the ones and zeros from a compact disk and then convert it into analog voltages. Non of those methods are 100 percent efficient and therefore losses occur. Cartridges are vulnerable to mechanical vibrations and feedback, tapes are noisy and A/D converters chop the signal into a sequence of bits and have a limited signal to noise ratio. Once the signal is converted to analog voltage it is being sent to the mixer, where it runs through various circuitry such as slider volume potentiometers, equalizer and filter circuits. Most of today s mixers are digital. They preform yet another A/D conversion. Most mixers have an input level and an output level that have certain voltage ranges. If ran outside those ranges, mixers 5

10 saturate and distort or limit and clip in the case of digital mixers. Filters and equalizers split the signal into bands and further amplify or damp it. This adds phase differences between these bands and could potentially phase out some of the frequencies. Once the sound is out of the mixer it most often goes to a DSP processor. DSP s are used to equalize or alter the dynamics of the signal. If used correctly they can improve the final result, but most incorrectly set up DSP s degrade the signal additionally. In order to protect the amplifiers and put a ceiling to the sound system because of health standard recommendations or regulations limiters are set up to limit the signal so that it doesn t exceed a certain value. Those kind of limiters are called brick wall limiters and to what their name suggests. They decrease the dynamic range by limiting the peaks, forcing the quiet parts of a song to become loud, but since the loud parts can t get any louder, they just become more compressed and end up having no dynamic variation at all. DSP s these days are mostly digital - some of them are 24 bit 94 khz which is a fairly good standard, but older ones are often 16 bit. They preform an additional A/D and then D/A conversion to the signal. When all the processing is done, the signal is fed to the input of the amplifier. Amplifiers as the name suggest amplify the signal power so that it can shake the speaker membranes in order to produce vibrations. These vibrations cause the air in front of the membranes to compress and decompress. Those waves of compression and decompression then see low impedance in face of free air and propagate in open space. 3.2 Perception Sensitivity The sensitivity of the human ear is not linear along the whole audible frequency range. It is most sensitive at the frequencies where human speech is located. Sensitivity rolls off exponentially at low frequencies down to 20 Hz. Frequencies below 50 Hz are generally perceived by one s body rather than ears. The shape of the hearing curve also changes with sound pressure. The louder the sound, the more linear the curve becomes, the quieter the sound, the more expressed the sensitivity difference at different frequencies is. Figure 3.2 shows the Equal-Loudness curve at different levels. Hearing degrades during the course of a person s life. Prolonged exposure to loud noises and aging damages the outer hair cells in the cochlea causing hearing loss. High frequency sensitivity gradually decreases with age as well as the ability to depict single sounds in a noisy environment. Therefore highly 6

11 Figure 3.2: Equal Loudness Curve reverberant places might sound normal to a young person, but will sound irritating for an old one. Dynamic Range Another concept that is important to this research is the dynamic range perception. In nature it varies from the noise a leaf makes when it strikes the ground, to thunderstorms and earthquakes. Dynamic range is the measure of the difference between the most quietest and loudest level in a sound piece. For over a decade producers, labels, mixing and mastering engineers have been pushing the limit of how loud a record should be in order to sell. Utilizing different dynamic range compression techniques, they have been able to reduce the dynamic range their songs to less than 5 db, meaning that out of the 90 db of available range for a Compact Disc, all that happens is squashed between the 0 and -5 db. In Figure 3.3 is Michael Jackson s Black or White in it s original master in 91 and the two consecutive masters in 95 and 07. 7

12 Figure 3.3: Black or White Amplitude Plot 3.3 Reverberation Reverberation is the sum of reflections from all surfaces of a room, when a sound source is introduced. The reverberation varies in time and frequency depending on the size, geometry and material of an enclosure (room). Figure 3.4: Watterfall Plot of Sound Decay in a Room In Figure 3.4 is a waterfall plot of the reverberation in a room, where each colored layer represents the frequency response of that particular time slice. Most rooms feature parallel surfaces: floor-ceiling and walls. These give rise to standing waves which tend to amplify the sound level at frequencies corresponding to the distance between the surfaces. The frequency of the standing waves can be expressed as f = λc, where λ is the wave length in meters and c is the speed of sound. Every time a sound wave bounces off a surface, its energy is reduced. The amount with which the energy is reduced depends on how well the surface 8

13 absorbs sound. The sound pressure level drops by 3 db for every doubling of the distance from the source as well. Since most materials don t reflect sound perfectly and sound power level decrease with distance, it can be assumed that sound will decay in time. Taking into account that the hearing sensitivity rolls of exponentially at low frequencies, and that sound decays in time, it can be concluded that at very low frequencies, reverberation is not perceived the same way as in higher frequencies in which the human ear is more sensitive. Studying reverberation is very important, because it is the blueprint of a room or space. The reverberation tail of a room gives great information about the room s character. Since human ears can pick up the direction from which the sound is coming, perceived reverberation gives information to the brain regarding what the size of the room is, what the materials inside are like as well as what the shape of the room is. 9

14 Chapter 4 Simulation In order to simulate the various acoustical properties, a vector ray tracing program is used. It generates random rays from the source to the receiver position, accounting for the time and amplitude change along the way. Then all the rays are summed for the received position, and an impulse response is generated. The vector ray tracing don t account for any phase differences, therefore it is not very reliable for low frequency simulation, although this is the case for having a trustworthy model corresponding to a real room. In this project this is not required, since all virtual room are hypothetical. Having generated an impulse response, an audio signal can be convolved with it in order to produce the sound as if it is played in the specific room. Figure 4.1: 3D Rendering Figure 4.1 shows a 3d rendering of the virtual room environment. The red spheres represent the two sound sources and the blue spheres represent 10

15 receivers. A 3d rendering of the room was implemented into the testing GUI in order to help participants orientate and ease them. The visual and auditory systems are tightly interconnected, so seeing a space gives additional input to the brain, thus helping it assimilate the environment the participant is located in. 4.1 Music Samples The music samples are selected to represent different sub-genres while still belonging to the same concept. Bellow is a chart of the samples used, the genre and release year. Nr. Artist Title Genere Year 1 Mario Basanov Closer Nu Disco Midland Placement Deep House Roman Fludegel Girls With Status House Symbols and Instruments Mood Techno 1989 During the test, the user had to compare between the same music sample, but in a different acoustical environment, so that any music preference can be omitted. Sample 1: Mario Bassanov Closer The first sample represents the electronic sub-genre of Nu-Disco that has been increasingly popular during the last two years. Figure 4.2: Original Sample Figure 4.3: Convolved Sample Figure 4.2 and Figure 4.3 show the spectral plot of the original music sample and its convolved version. On the spectrogram plot of the original sample, it can be seen that the original sample contains frequencies up to 20 khz and has most of its energy concentrated in the region bellow

16 Hz. However there is another concentration of high energy slightly above 1 khz which judging from the spectrogram is caused by the snare drum. On the spectrogram plot of the convolved sample, the effect of the room can be seen. The modes of the room amplify and sustain certain frequencies while damping others. For example frequencies below 50 Hz have been amplified while the 100 Hz band is being damped. Room damping patterns can be observed around 600 Hz and 6 khz. Sample 2: Midland Placement The second music samples represents the sub-genre of Deep House. It is a major part of the electronic music scene. Figure 4.4: Original Sample Figure 4.5: Convolved Sample Figure 4.4 and Figure 4.5 show the spectral plots of the original sample 2 and its convolved version. On the spectrogram plot of the original sample it can be seen that the sample contains frequencies from 20 Hz to 20 khz. The energy seems to be distributed mostly in the low frequencies, and rolling off slightly in the highs. On the on the convolved samples spectral plot, the effect of the rooms response can be seen. This room also amplifies the frequencies bellow 50 Hz, and seems to dampen the frequencies around 70 Hz. The midrange is attenuated to a certain extent too. The same pattern as in the other room is also observed here. Its the severe dampening around 6 khz. Sample 3: Roman Fludegel Girls With Status The third sample is categorized as house. Figure 4.6 and Figure 4.7 show the original and convolved version of sample number 3. On the original sample once again it can be seen that the sample contains frequencies representing the whole audible range. There is a very pronounced bass line that spans from 20 Hz until 50 Hz and contains 12

17 Figure 4.6: Original Sample Figure 4.7: Convolved Sample most energy. Also a chord in the frequency between 50 Hz and 100 Hz is seen. The kick drum from the second part of the sample introduces an energy balance change probably caused by the use of parallel or side chain compression. It takes away from the energy of the bass and introduces new peaks in the range from 20 to 500 Hz. The start of the hi-hat extends the frequency range up to 20 khz as well. In the convolved sample it can be seen that the room modes sustain the energy in between 20 Hz and 50 Hz while damping the frequencies around 70 Hz. The effect of the kick is less noticeable because of the strong damping in the range between 70 Hz and 500 Hz. The same pattern as in the previous samples can be seen in the region around 6 khz. Sample 4: Symbols and Instruments Mood The last sample represents the genre of Techno and is a classic example. It was released in 1989 and is produced and recorded with purely analog equipment such as Synthesizers, Drum Machines and Sequencers. Figure 4.8: Original Sample Figure 4.9: Convolved Sample Figure 4.8 and Figure 4.9 show the original and convolved version of sample number 4. On the last spectral plot it is seen that the bass line in the first part of the sample produces the most energy, located mainly in the lower 50 Hz, however it extends all the way up to 4 khz as well. After the 13

18 introduction of the beat, the frequency range extends all the way to 20 khz. From the spectral plot of the convolved sample it is seen that the energy in the range between 20 Hz and 50 Hz is generally sustained, while 60 Hz to 70 Hz band is heavily damped.the same phenomenon still exists as in the other rooms where the 6 khz band is being damped as well. 4.2 Default Room The first virtual room features characteristic materials that are used in building construction. Table 4.1 shows the dimensions and corresponding modes, due to the parallel construction of the room. Length Width Height m m 5.94 m Table 4.1: Default Room Modes Figure 4.1 shows a table of the first 17 modes in the original room. The absorption coefficients for the materials used in 1 octave band can be seen in Table 4.2 In the default room, a typical choice of materials was made in order to illustrate how an average room would sound like. The resulting Early decay time of this setup is shown in Figure The EDT plot shows how fast the sound energy in the room drops by 10 db in 1 octave bands. It is interested to notice that the 125 Hz band takes 1 14

19 Surf./Freq. 125 Hz 250 Hz 500 Hz 1 khz 2 khz 4 khz 8 khz Floor Ceiling Side Walls Front Wall Back Wall Panels DJ Booth Table 4.2: Absorption Coefficients Table 1 EDT 0.8 Time [s] Figure 4.10: EDT of Default Room second to decay, we can also later on see in Figure 4.12 that this prolonged decay results in a decrease of the clarity for that band. This can be explained by the low absorption coefficients of the materials as well as with the density of room modes in this frequency band. When looking at the T30 plot of the reverberation, it can be noted that 30 db from the instant acoustical energy, the absorption of the materials dominates over the modes and the reverberation times even out in the different frequency bands. That is not the case in the 125 Hz band though. There the sound takes over 1 second to decay, which is twice the time it takes the other bands. Figure 4.12 shows what was suggested from the EDT and T30 plots. The problematic 125 Hz band shows a lower clarity rating. The fact that most materials as well as the air absorb frequencies in the 8 khz band automati- 15

20 1 T Time [s] Figure 4.11: T30 of Default Room 1 C Clarity [db] Figure 4.12: C50 of Default Room cally leads to better clarity. The T30 plot seems to be inversely related to the C50 plot. So the longer the T30, the smaller the clarity rating for a specific band. Figure 4.13 is the Spectral Plot of the impulse response of the Default room. It combines the information from the times and frequency domains in one place. It makes it easy to distinguish certain features of the signal such as standing waves adding in phase and out of phase. From the default room impulse response spectral plot it can be seen that the most energy preserved in the room is in the low frequencies. There is a big collection of standing 16

21 Figure 4.13: Spectral Plot of Default Room waves below 50 Hz. Going up in frequency there is a modal overlap at around 150 Hz. There is a prolonged decay at 2 khz too. The sound decays quite fast in the 1.5 khz region. 4.3 Fixed Low End Reverberation In this room which is analogous to the Default room, the absorption coefficients of the 125 Hz band for all materials (except for the DJ booth) have been increased to 0.8. This is done in attempt to illustrate the effect of the low frequency reverberation i terms of overall sound perception. Surf./Freq. 125 Hz 250 Hz 500 Hz 1 khz 2 khz 4 khz 8 khz Floor Ceiling Side Walls Front Wall Back Wall Panels DJ Booth Table 4.3: Absorption Coefficients Table All materials absorption coefficients have been kept the same as the Default room, apart from the 125 Hz band coefficients which have been increased 17

22 in order to tame the low end reverberation. 1 EDT 0.8 Time [s] Figure 4.14: EDT of Fixed Low End Reverberation Room Already in the EDT plot, there is an improvement, the 125 Hz band taking as little as 0.6 seconds to decay by 10 db. This change makes the room feel much more balanced. However achieving such a low EDT in real conditions is hard and costly. 1 T Time [s] Figure 4.15: T30 of Fixed Low End Reverberation Room The T30 plot of the Fixed Low End Reverberation room shows an even better handling of the low end absorption. Bringing the time required for 18

23 the late sound decay to -30 db to under 0.5 seconds. The T30 plot seems almost flat in all other frequency bands. 1 C Clarity [db] Figure 4.16: C50 of Fixed Low End Reverberation Room As predicted by the T30 response, the Clarity index seems to be spread evenly across all frequency bands. Figure 4.17: Spectral Plot of Fixed Low End Reverberation Room On the spectral plot for the low end reverberation room it can clearly be seen that the low end reverberation is significantly reduced and the room has a much more linear fashion in terms of decay time in the different frequency ranges. This change improves the overall perception. 19

24 4.4 Hearing Curve Room In this modification of the Default room, the material absorption coefficients have been changed in order to achieve a reverberation curve, that is comparable by the human equal loudness curve. Meaning that frequencies that the ear is more sensitive to will take longer to decay and frequencies that the ear is less sensitive to are going to take less. Surf./Freq. 125 Hz 250 Hz 500 Hz 1 khz 2 khz 4 khz 8 khz Floor Ceiling Side Walls Front Wall Back Wall Panels DJ Booth Table 4.4: absorption Coefficients Table The absorption coefficients have been modified in order to achieve the desired reverberation curve, but the geometry of the room has not been changed. 1 EDT 0.8 Time [s] Figure 4.18: EDT of Hearing Curve Room In the EDT plot, it is clear that the room has a very uneven decay character in the different bands. The room was designed to match the reverberation 20

25 to the hearing sensitivity as a primary objective and this is how the EDT corresponds. It could be used when comparing in the statistics part. 1 T Time [s] Figure 4.19: T30 of Hearing Curve Room The T30 of the Hearing Curve Room relates directly to the human hearing sensitivity. It can be seen that the mid and high frequencies to which the ear is very sensitive have a prolonged decay time. The general reverberation of the room is also longer compared to the Default Room. 1 C Clarity [db] Figure 4.20: C50 of Hearing Curve Room The Clarity plot is in a direct relation with the T30 plot, and shows decreased index in the 125 Hz band as well as in the very high frequencies. 21

26 The high reverberation in the 8 khz band is hard to be achieved as well, since the air absorbs a lot of it, as well as most materials in a typical room. Another thing to be noted is that the scattering for frequencies above 4000 Hz is very high which also helps quicken the decay. Figure 4.21: Spectral Plot of Hearing Curve Room Figure 4.21 shows the spectral plot of the impulse response of the hearing curve room. On the plot it can be seen that the most energy is contained in the low end. There being a high modal overlap in the frequencies below 50 Hz. Some single modes in the range between 50 Hz and 100 Hz can be identified too. A modal overlap can also be seen just above 100 Hz. In this room, the decay follows the usual tendency, but then at 10 khz, the decay time increases again. This is due to the fact that the materials selected in the room, reflect the frequencies that the ear is not so sensitive to, and as the human sensitivity rolls off pretty fast after 8 khz, the reverberation times increase in order to compensate. 22

27 4.5 Inverse Hearing Curve Room In the Inverse Hearing Curve Room, the T30 is inversely proportional to the human hearing curve. This means that frequencies that the human hearing system is sensitive to are decaying fast and frequencies that the ear is not sensitive to have a prolonged decay. In a way this will theoretically even out the perceived reverberation. Surf./Freq. 125 Hz 250 Hz 500 Hz 1 khz 2 khz 4 khz 8 khz Floor Ceiling Side Walls Front Wall Back Wall Panels DJ Booth Table 4.5: Absorption Coefficients Table The absorption coefficients were modified in such a way as to produce the required T30 decay curve to match the inversely proportional human hearing curve. 1 EDT 0.8 Time [s] Figure 4.22: EDT of Inverse Hearing Curve Room From the early decay curve, it can be seen that the 125 Hz is quite pronounced in terms of reverberation time as well as the 2 khz and 4 khz bands. 23

28 1 T Time [s] Figure 4.23: T30 of Inverse Hearing Curve Room Looking at the T30 plot it is clearly different from the EDT. The late reflection is much more subtle and even in the different frequency bands. The 125 Hz band though takes almost 1 second to drop by 30 db. However this is what the inverse hearing curve suggests. 1 C Clarity [db] Figure 4.24: C50 of Inverse Hearing Curve Room Taking a look at the Clarity, it is very evenly distributed through the different frequency bands, keeping a high value at the same time. Theoretically this is a good sounding room. Figure 4.25 shows the spectral plot of the inverse hearing room. From the 24

29 Figure 4.25: Spectral Plot of Inverse Hearing Curve Room plot can be seen that the most energy is still contained in the low frequencies. Again there is a high modal overlap below 50 Hz as well as around 100 Hz. Some of the modes that were excited in the hearing curve room doesn t seem to be excited in this one. As the frequency goes higher the decay time decreases, which is a more normal behavior for a rooms reverberation. 25

30 4.6 Reflective Side Wall Room In this room, the front and back walls are made of a material that is completely absorptive at all frequencies. This room is made in order to test the spaciousness perception. Surf./Freq. 125 Hz 250 Hz 500 Hz 1 khz 2 khz 4 khz 8 khz Floor Ceiling Side Walls Front Wall Back Wall Panels DJ Booth Table 4.6: absorption Coefficients Table All materials are the same as in the Default room, except for the front and back walls, which now are made of a totally absorptive material. 1 EDT 0.8 Time [s] Figure 4.26: EDT of Reflective Sides Room The EDT curve is very similar to the Default room curve, this is so, because of all the materials being the same, apart from the absorptive front and back walls. Since their absorption area is not that big, they don t introduce that big of a change. 26

31 1 T Time [s] Figure 4.27: T30 of Reflective Sides Room The T30 is very similar to the Default room too, with the difference, that since 2 of the walls are completely absorptive, the late reflections will decay at a higher rate. 1 C Clarity [db] Figure 4.28: C50 of Reflective Sides Room The Clarity factor index is quite flat and very high across the whole spectrum, with a little dip at 125 Hz. Theoretically this should be a very good room. Figure 4.29 shows the spectral plot of the impulse response of the room with reflective side walls. From the plot it can be seen that there is an 27

32 Figure 4.29: Spectral Plot of Reflective Sides Room increased modal density in the frequencies below 40 Hz as well as slightly above 100 Hz. The reverberation time decreases with frequency. 28

33 4.7 Reflective Front and Back Wall Room In this room, the absorption coefficients are the same as in the Default room, with the difference that the side walls are totally absorptive. This would make the room produce a narrower sound image. Surf./Freq. 125 Hz 250 Hz 500 Hz 1 khz 2 khz 4 khz 8 khz Floor Ceiling Side Walls Front Wall Back Wall Panels DJ Booth Table 4.7: absorption Coefficients Table 1 EDT 0.8 Time [s] Figure 4.30: EDT of Reflective Front and Back Wall Room From the EDT it can be seen that the first order reflections decay quite fast. This is due to the fact that the side walls represent a big percentage of the absorption area. However the 125 Hz band requires almost 0.8 seconds to decay 30 db, which is due to the room geometry. The T30 fluctuates around 0.5 seconds for all frequency bands, except for the 125 Hz. This is significantly drier than the rest of the rooms. 29

34 1 T Time [s] Figure 4.31: T30 of Reflective Front and Back Wall Room 1 C Clarity [db] Figure 4.32: C50 of Reflective Front and Back Wall Room As expected, the dryness of the room, calls for a higher Clarity rating, as it is inverse proportional to the T30. However, in the 125 Hz band, the rating is not that high. Figure 4.33 shows the spectral plot of the impulse response of the room with reflective front and back walls. As seen from the plot, most of the energy is contained in the low frequencies. There is a high modal overlap in the frequencies below 40 Hz. At 125 Hz there is a high density of modes too which boosts those frequencies additionally. However it does not affect the reverberation time significantly. The decay time decreases with frequency. 30

35 Figure 4.33: Spectral Plot of Front and Back Wall Room Exceptions are the 250 Hz and 2 khz regions where the decay is slightly shorter. 31

36 4.8 Concrete Room For the sake of illustration, a model of a room made entirely out of concrete is created. However extreme this case might seem, most venues actually are closer to concrete rooms, rather than listening rooms. Surf./Freq. 125 Hz 250 Hz 500 Hz 1 khz 2 khz 4 khz 8 khz Floor Ceiling Side Walls Front Wall Back Wall Panels DJ Booth Table 4.8: Absorption Coefficients Table All room walls, ceiling and floor are made of concrete in this extreme case. This room is very useful in finding specific features created by the geometry, since the walls do not absorb. Therefore all modes will add up and exaggerate the modal pattern. The DJ booth remains the same as in the Default room. 6 EDT 5 4 Time [s] Figure 4.34: EDT of Concrete Room The EDT curve shows a very constant decay time decrease rate starting from 250 Hz until 8 khz. The 125 Hz band s decay time seems to not be very long compared to the overall decay. 32

37 6 T Time [s] Figure 4.35: T30 of Concrete Room The T30 and EDT plots look the same, this is due to the fact that since there is no real absorption in the room, the early and late reflections have a similar pattern. They mostly represent a sum of all the room modes. 1 C Clarity [db] Figure 4.36: C50 of Concrete Room As expected, the Clarity index is very poor, but tends to improve with high frequency, since the concrete has some absorption in the high end. The clarity index being a rating for the difference between the sound energy in the first 80 milliseconds, is it expectedly low since the overall decay takes a very long time. 33

38 Figure 4.37: Spectral Plot of Concrete Room Figure 4.37 shows the spectral plot of the impulse response of the concrete room. From the plot it can be seen that the decay time of this room is very long. It is a very useful room for finding out patterns in the decay caused by the geometry of the room alone. Most of the energy is being preserved in the frequencies below 1000 Hz. The modal overlaps can be seen easily and their effect is distinguishable. There is a prolonged decay in the 125 Hz region as well as in the 400 Hz and 1 khz and 2.5 Khz. Those peaks are caused by the geometry. 34

39 Chapter 5 Processing In this chapter, the various methods and techniques used for obtaining and processing data are discussed. Deconvolution is used in order to artificially create an impulse response from a sine sweep reproduced by a speaker (in real life) or by an array of vectors with a different frequency contend (in the simulations) Convolution is later on used in order to examine the way a music piece will sound in a a room. In the measurements subsection, the music sample properties will be discussed. 5.1 Deconvolution The deconvolution is an algorithm based process used to typically reverse the effects of convolution. In this case, the effects of convolution being the natural reverberation of a room. Whenever a sound is played in a room, it is being reflected from all walls, leaving a trail of decay. Using a sinusoidal sweep signal, a speaker can be excited in the whole audible range, and then by deconvolution, the time difference between different frequency content can be omitted, leaving only the impulse response for the whole audible range. This is very handy, because it can be used to analyze the properties of a room, in the same way impulse responses are used to inspect electronic circuits. 5.2 Convolution Once the impulse response of a room is captured, it can be used to compute how music will sound in the room. This is done by multiplying each sample of the music piece with each sample of the impulse response. This way every sample of the music piece gets it s decay trail, that is being added to the next sample and so on. 35

40 5.3 Measurements In this section, the sound samples are passed through a signal analyzer. The phase difference and frequency response are being measured. The phase meter shows what the phase difference between the two stereo channels is to show how wide or narrow the sound image is. If two pure tones with the same phase are played, the graph will show a narrow line in the center meaning the relation is 1, or the sound is mono. If the two samples are phased 90 degrees relative to each other, there will be a horizontal line, showing that the relation between the two is 0 or the samples are inversely proportional. When heard through headphones such sound will be disturbing. When played through speakers it should theoretically cancel out at some frequencies, causing it to sound out of phase Question 1 In question one the sound samples of the room with reflective side walls is compared to the one with reflective front and back walls. Figure 5.1: Front and Back Figure 5.2: Sides Under careful examination it can be noticed that the room with reflective side walls shows an image that is slightly wider, and also shifted towards the left. This is due to the fact that the listening position is off center, therefore the sound reflected from the near wall will contribute to the sum resulting in a higher amplitude. The frequency responses of the two rooms seem identical apart from two small differences. The first one is small bump slightly below 250 Hz in Figure 5.4 and absence of the peak at 100 Hz that is observed in Figure Question 2 In the second question the sound samples of the room with a T30 response that follows the hearing curve is compared with the room with a T30 curve 36

41 Figure 5.3: Front and Back Figure 5.4: Sides resembling a curve that is inverse proportional to the hearing curve. Figure 5.5: Hearing Curve Figure 5.6: Inverse Hearing Curve Comparing Figure 5.5 and Figure 5.6 it can be seen that the Inverse Hearing Curve room has a wider image which is caused by the generally longer reverberation. Figure 5.7: Hearing Curve Figure 5.8: Inverse Hearing Curve In the frequency response comparison, Figure 5.8 has higher levels in the midrange, caused by the longer reverberation which conserves the sound energy Question 3 In the third question, the default room reverberation is used to test the difference between an original music sample and the compressed version of the same sample that is convolved with the frequency response of an iphone headphone. 37

42 Figure 5.9: Original Sample Figure 5.10: iphone Emulation In the phase plot, the compression is clearly visible, however providing the same room, a louder sound will produce a longer tail. Figure 5.11: Original Sample Figure 5.12: iphone Emulation The frequency response comparison shows that the iphone sample is clearly louder at the mid and high frequencies. The low frequencies though roll off much faster below 60 Hz Question 4 The final forth question compares the audio sample convolved with the impulse response of the default room to the same sample convolved with the impulse response of the fixed low end reverberation room. Figure 5.13: Default Room Figure 5.14: Low Frequency Fix 38

43 Figure 5.13 seems to be generally louder and have a bigger spread especially in the anti phase area. This could be a good sign if the off phase content has high frequency character, or a bad one if it is low frequencies. Figure 5.15: Default Room Figure 5.16: Low Frequency Fix Comparing the frequency response of the two rooms, the only noticeable difference is the increased amplitude of the 250 Hz and 500 Hz band. in the Default room. The low frequency fix room has an even extended low frequency range. Note: Wider sound image does not necessarily mean better sound. More spread in high frequencies is a desired effect, as opposed to spread in low frequencies. Since low frequency sound tends to be radiated omnidirectionally, out of phase bass can cause trouble. It will cancel out at some frequencies and sound disturbing at others. 5.4 Testing Program A graphical user interface was designed in Matlab s GUIDE editor. The GUI consists of an image - a 3d rendering of the person s location in the virtual room that helps the participants orientation. The buttons are located in the lower part of the interface. There are 4 sections, corresponding to the 4 questions. Each section has two options, with two buttons each. One for playing the sounds and one for choosing. In the bottom there is a submit button, that completes the process and advances the database to a new participant. The GUIDE editor allows graphically build the graphical user interface s layout and then translate it into matlab code. The play button utilizes the soundsc function in Matlab that scales the convolved vector to normalized audio levels, sot that it can be played without 39

44 Figure 5.17: GUIDE GUI editor Figure 5.18: Play Button Callback Function distortion. Once the sound is started, the playback cannot be stopped until the sample is over. This is a drawback, since the user might press two buttons at the same time causing the sound samples to overlap, but it is the only option in this layout, since the callback option s lifespan is limited to the button click. After that the variables are being deleted, so the play function would not work. The parameters supplied to the soundsc function are the sampling frequency and the vector, which contains the two stereo channels. 40

45 Figure 5.19: Choose Button Callback Function The choice button enables participants selection to be recorded. Again since the button callback function s limitation, the database where the statistics are recorded is being opened, data is being written to it and then saved. The variable which selects which entry of the DB is being written is stored in the database, and is being incremented only when the submit button is pressed. 41

46 Chapter 6 Statistics Using the convolved music samples, different combinations of room setups and audio samples were used in order to study the effect of room acoustic as a tool to enhance audible experience. The goal being to find a relationship between preferred reverberation times at different frequencies and the relationship between early and late reflections. A statistics experiment was also performed testing a number of participants on their hearing curve. Figure 6.1: Experiment GUI 42

47 In Figure 6.1 shows the graphical user interface of the program used to perform the experiment. There are 4 test questions that have 2 options each. The user was given the task to pick the better of the two for each question. These questions were chosen specifically, in order to gather specific information on people s preference. More details can be seen the description for the individual questions. Choice A Choice B Question 1 Reflective Front/Back Reflective Sides Question 2 Hearing Curve Inverse Hearing Curve Question 3 Original Sample Compressed iphone Transfer Function Question 4 Default Room Fixed Low End Reverberation Table 6.1: GUI answers In order to ease the participants, and avoid listening fatigue, the experiment was designed to be as short as possible. The sound samples played are 8 seconds long and can be played and replayed at any time for easy comparison. Playing many different audio samples can confuse people and therefore reduce the fidelity of the experiment. Comparison between two samples gives less room for confusion and is less tiring for the participants. The participants were chosen to represent a wide range of ages and occupations. An effort was made to keep the test group as diverse as possible. Participants age varies from 18 to 55 years, the gender and occupation are diverse. However there is a significant amount of students taking part. Occupations of participants vary from construction workers to business consultants. People who attend loud concerts and others who live a quiet life, people who have noisy a work environment such as carpenters and people who spend their time in an office. In Figure 6.2 the choice bias for all 4 questions is showed, where a score of 0 represents choice A and 100 represents choice B. Since this is a plot of the average answer, it is biased between the two extremes to show the choice as a percentage. The outcome of the experiment does not confirm nor reject the theory used to create the questions, but rather shows a tendency in people s preference. This is important since the end user is the one to experience. A parallel statistical experiment was performed in order to measure on the hearing sensitivity of the participants. A number of participants were tested in this stage, in order to confirm if the theoretical concepts fit the measurement. 43

48 Mean Choice [%] Mean Choice Question Number [Hz] Figure 6.2: Mean Score of Participants Oppinion 40 Participants Sensitivity Curve Relative Sensitivity [db] Figure 6.3: Curve of participants average sensitivity Figure 6.3 shows a shape very similar to Figure 3.2. The sensitivity gradually rolls of for frequencies below 250 Hz. The same happens for the frequencies above 8 khz, but in this case the slope is much bigger. Figure 6.4 shows the hearing curves of participants representing different age groups. From the plot it can easily be seen that there is a tendency for the high frequencies to roll off as people age. This is due to the fact that the tiny hairs in the cochlea get worn out and damaged during the course of a person s life. All mammals loose their hearing progressively, and it does no regenerate. However new stem cell research is breaking through hearing recovery, but it it in an early stage yet. Figure 6.5 shows the hearing curves for participants with a different occupation. To a certain extent the occupation of a person determines how fast their hearing will degrade. Exposure to loud noises for an extended period 44

49 Normalized Sensitivity [db] Hearing Sensitivity Curve by Age 18 years 35 years 55 years Figure 6.4: Sensitivity Curve For Different Ages Normalized Sensitivity [db] Hearing Sensitivity Curve by Occupation DJ Student Carpenter Figure 6.5: Sensitivity Curve For Different Occupations of time damages the hearing system in an irreversible way. There are regulations regarding noise at the work place, but yet carpenters, heavy machine operators and construction workers are spending most of their work time in a noisy environment. There is a statistic showing that very short impulse noises damage the hear more than constant noise. The instant noise damages the physical part of the system, while the constant noise is more affecting the brain and neural part of the hearing system. 6.1 Question 1 Front and Back Reflection vs. Side Reflection Question 1 was designed to test on the perceived quality in regards of 45