Proceedings of the Institute of Acoustics

 

Standing on the shoulders of giants

 

G. Leembruggen FIOA, Acoustic Directions, ICE Design Australia

 

1 INTRODUCTION

 

This paper expands on the lecture I gave to the Reproduced Sound 2022 conference after receiving the Peter Barnett Memorial Award for 2022.

 

My career in electro-acoustics and acoustics spans some 40 years, and I have been fortunate to have worked on many varied, highly interesting and challenging projects. In many situations, the most suitable solution has been something different to the norm; such as development of a bespoke loudspeaker to solve an acoustic problem, construction of a tool, or using an existing product in a new way. But this approach to projects often involves stepping into unfamiliar territory and my foundation for these endeavours has been the work of people I regard as “giants”.

 

1.1 This Paper

 

This paper covers such topics as:

• People who have inspired me to explore solutions. I consider these people to be giants in the electro-acoustic and acoustic realms, and interestingly, many of them are previous recipients of the Peter Barnett Award.
• Tools I have built to aid analysis and measurement.
• Some of my design that employ bespoke hardware.
• Aspects of certain projects that didn’t work as I expected.

 

As this paper is a collection of (hopefully) interesting but disconnected topics, it may seem a little disjointed.

 

Constant themes in my work have been the desire to produce a direct sound field with great fidelity for listeners, and to minimise late-arriving sound at all frequencies. In essence, my holy grail has been to deliver constant directivity, intimate sound and a flat frequency response.

 

In the strict measurement domains of speech intelligibility, the term “intimacy” has no meaning. But I have found that the more intimate the sound is, the easier it is to reliably understand speech under the wide range of operational circumstances and the more comfortable speech is to listen to. Accordingly, it serves as an indicator of subjective intelligibility.

 

All of the sound system projects I have been involved in since 2003 were undertaken in collaboration with David Gilfillan. Mark Hanson and I designed the acoustics of the Supreme Court of NZ and the National Library auditorium. The benefits of long-term collaboration with David and Mark have been profound.

 

2 MY BEGINNING

 

2.1 Dr Richard Small

 

Dr Richard Small from the Electrical Engineering Faculty at Sydney University in 1976 was responsible for me commencing my electro-acoustic career. At an end-of-final-year student function, Dick Small asked me if I would like to come into the university and measure some loudspeakers. I agreed and he encouraged my learning and became a technical mentor. I loved the science of reproducing sound.

 

Dick once told me that I would need to continually build my own tools, and this has proven true.

 

Dick also drew inspiration from Dr Leo Beranek’s book Acoustics [1] and it has been a constant companion for me.

 

2.2 Neville Thiele

 

Dick Small considered that his work stood on the shoulders of Neville Thiele, who has also been a giant for me.

 

Besides his work with vented-box loudspeakers, another of Neville Thiele’s skills was his knowledge of passive electronic circuits. Neville had an intuitive understanding of passive filters and impedance compensation networks, which greatly assisted my design of bespoke loudspeakers for the Australian Parliament and the High Court of Australia.

 

 

2.3 Siegfried Linkwitz

 

In the late 1970s, Linkwitz [2] wrote a series of articles on loudspeaker system design in Wireless World magazine. I devoured these articles as they were written in a tutorial style. An example of his articles is shown in Figure 1 below.

 

 

Figure 1: Example of the tutorial-style article by Linkwitz.

 

3 MEASUREMENT AND DESIGN TOOLS

 

There are five measurement and design tools that have been of great assistance to my work. These tools are described below.

 

3.1 Half-Space Measurement Table

• Inspired by the half-space measurement platform that Dick Small installed at Sydney University in 1976, I constructed a half-space table in my home lab. to enable us to measure the frequency responses of loudspeakers in a half space environment.
• The table retracted to form a storage box for our test enclosures to measure the Thiele-Small parameters of loudspeaker drivers.
• Figure 2 shows a prototype loudspeaker sitting on the retracted half-space table. The loudspeaker comprised a dual 380 mmm driver with a high frequency horn and was intended for commercial sale. This system used frequency shading to minimise spatial aliasing due to the large distance between the centres of the two low/mid drivers. retracted half-space table.
• We used the first IQS 401 FFT analyser in Australia (1984) to measure the responses of drivers in a half space situation. The analyser was a card that was located inside an Apple 2e computer.

 

 

Figure 2: Prototype loudspeaker sitting on the retracted half-space table.

 

3.2 Measurements of Short Reverberation Times in One-Third Octave Bands

 

In 1984 inexpensive acoustic analysers were not yet available. The problem was how to measure reverberation times of less than 200 ms in one-third octave bands, given that 1/3rd octave filters have their own decay time.

 

The solution was to invert the problem and excite the room using a Hanning-shaped toneburst signal with a bandwidth of 1/3^rd octave as per Linkwitz [3]. The resulting decays of sound could then be captured on a storage oscilloscope.

 

The plots in Figure 3 compare the spectra of tonebursts with rectangular and Hanning windows.

 

However, as I didn’t have a storage oscilloscope reading dB and I couldn’t’ afford a TEF (TimeEnergyFrequency) analyser, I designed a time-gating amplifier with a variable time gate and gain to allow the IQS 401 FFT (Fast Fourier Transform) analyser to show later parts of the decay curve with higher amplitudes.

 

Figure 3: Spectra of tonebursts shaped with rectangular and Hanning windows.

 

3.3 Prediction of Sound Absorption Behaviour

 

We needed to design bespoke absorbers for auditoria, studio and control room projects and in 2002 the software tools to predict sound absorption were not readily available.

 

I developed an Excel spreadsheet using calculations of forward and backward travelling waves with various impedance layers. These calculations can be also implemented using the so-called transfer matrix method. Stuart Colam and I prepared a paper [4] for RS19 describing this prediction method.

 

The conceptual diagram in Figure 4 shows the alternating systems of solid impedance structures and air spaces that could contain insulation of some form.

 

Later, I included data for polyester insulation from Garai & Pompoli [5] and fibreglass insulation from other researchers to predict their characteristic impedances and complex propagation constants. This spreadsheet has proven extremely useful over the years for room acoustic design. Only recently have similar software packages become available.

 

Figure 4: Conceptual arrangement of acoustic elements in the calculation.

 

3.4 Acoustic Impedance Tube

 

A manufacturer of polyester insulation wanted to understand the sound absorption properties of various densities and fibre structures of their insulation. As tests in reverberation chambers were too expensive for this type of development exercise, the best way was to measure the absorption in an impedance tube.

 

As we couldn’t afford a Bruel and Kjaer impedance tube, pictured at right, I elected to build one.

 

Inspired by a paper by Vanderkooy and Stevens [6], I built a 6 m long impedance tube, which uses a single microphone located in the middle of the tube to measure the impulse response (IR) with the sample at the end of the tube. The incident and reflected waveforms can be readily seen in the impulse response and isolated. From the amplitude and phase responses of these two waveforms, the sound absorption properties of the material under test can be calculated. In 2021, I extended the tube out to 9 m.

 

 

Figure 5: Bruel and Kjaer impedance tube.

 

The inside diameter of the tube is 84 mm, yielding an upper frequency limit of around 2.2 kHz. With its 9 m length, we can measure down to 50 Hz, which is lower than commercial impedance tubes.

 

Figure 6: Left to right. Sections of the low and high frequency impedance tubes; sample holder with fabrics over foam backing; inner section of the low-frequency tube with the microphone and high frequency tube.

 

In conjunction with David Gilfillan, I prepared a paper for RS2010 [7] comparing the measured (with this tube) sound absorption coefficients of a range of Helmholtz absorbers backed by insulation with those predicted using our travelling wave spreadsheet. The agreement between measured and predicted sound absorptions was very encouraging.

 

To enable the measurement of the sound transmission loss of cinema screens at high frequencies [8], I built a high frequency tube in 2012 with an internal diameter of 19 mm to allow measurements up to 10 kHz.

 

3.5 Measurement of Flow Resistance

 

When designing critical listening spaces and environments where speech intelligibility is important, it is vital to accurately model sound absorption systems, and to do this requires data for the flow resistances of porous elements such as fabric and insulation. As flow resistance data is rarely published for fabric and insulation, we need to measure it. Another factor is that fabrics liked by architects mostly have higher flow resistances placing them on the edge of unsuitability.

 

As we couldn’t afford a commercial measurement unit, I elected to build a flow resistance meter based on ANSI standard C522-03 (2016), (shown in Figure 7).

 

A description of the meter was prepared for RS2020 [9], in which we noted that plots of flow resistance versus air-flow rate showed that the resistance varied with flow rate (see Figure 8).

 

We subsequently investigated this problem and found that when the pressure versus flow rate was plotted, the relationship between these two parameters was highly linear, indicating constant flow resistance with flow rate. (See Figure 9)

 

Offsets in the measurement of flow-rate and pressure due to tolerances expressed as a percentage of full scale reading are believed produce the non-linear plots of resistance versus airflow. This was an embarrassing “rookie” error.

 

With the purchase of a second flow-rate meter for very low flow rates, we can now measure flow rates over a range of 100:1.

 

 

Figure 7: View of flow-resistance apparatus

 

 

Figure 8: Example of flow resistance changing with flow rate

 

 

Figure 9: Examples of constant flow resistance with flow rate over the range 0.24 L/min to 24 L/min

 

3.6 Image Sources: The Effects of Room Boundaries

 

This topic is not a tool as such, but is a concept I have been ever mindful of when positioning loudspeakers close to reflecting planes.

 

One factor that can badly damage the direct field of a loudspeaker is the presence of image sources that are sufficiently close to the speaker to produce deep interference nulls with the direct sound.

 

The effects of image sources in room boundaries were first documented by Waterhouse, followed by Roy Alison [10]. Glyn Adams [11] integrated the image sources into a room with modes and calculated the frequency responses of a perfect low frequency loudspeaker at various times during development of the room modes.

 

Graphs from Adam’s paper are compiled in Figure 10 and show the sound power responses of a loudspeaker located at x =y = z = 1 m in a room that is 6 m x 5 m x 3.5 m at four time periods.

 

The graph shows that the primary drivers of the in-room power frequency response are the image sources which is simply overlaid with the effects of room resonances. Image sources that have affected my work are discussed later in this paper.

 

From my observations of today’s audio practice, I conclude that image sources are rarely considered by audio practitioners.

 

 

Figure 10: Frequency response at various times of ideal loudspeaker with image sources in a room.

 

4 PROJECT WORK

 

4.1 Low Frequency Analysis & Thiele Small Parameters

 

One of the first loudspeakers I designed was in 1983, which used two KEF B110 125 mm drivers and a Peerless D26 dome tweeter mounted in a 36-litre enclosure with a D’Apollito configuration. This configuration was actually first used in 1976 by Dr Bob Frater of Sydney University.

 

Underdamping each B110 driver using a series resistor of 3 ohms (termed Rg) allowed the system to achieve a 4th Order Butterworth response with a - 3 dB point of 42 Hz and a respectable system sensitivity of 88 dB for 2.83 volts; this performance was almost unheard of in a relatively small system.

 

 

We sold about 20 of these speakers but eventually realised that the world does not beat a path to your door, without a marketing department!

 

Equations and methods described in various papers by Dick Small [12], [13], [14] were used to predict the performance of the system and verify its performance. Examples of the system performance are shown in Figure 11.
 

 

Figure 11: Predicted frequency responses, cone excursion and displacement limited SPL for the dual B110 system.

 

4.2 High Court of Australia – Near-field Array Loudspeaker

 

Coverage for two rows of the public gallery in the High Court could only be provided using loudspeakers located on a concrete barrier directly behind the listeners. Only a single loudspeaker feed was available.

 

The solution [15] uses 34 x 65 mm drivers in an array located on a pelmet above and behind the listeners. The drivers are arranged in a curved and delayed format, to minimise the phase interference effects. Passive all-pass filters were used to delay the outer drivers that could not be located on the curved baffle. Figure 12 shows the nearfield array, along with our measurement microphone.

 

 

Figure 12: View of nearfield array above gallery seats in the High Court.

 

Neville Thiele generated the transfer functions for the passive all-pass filters, as he understood the mutual coupling between inductors which are used in bridged-T all-pass networks. His work removed the need for coupled inductors in the array’s filters. The passive network comprising equalisers and all-pass filters is shown in Figure 13.

 

 

Figure 13: Passive equaliser and all-pass filter network used in the near-field array.

 

5 BEAM-CONTROLLED ARRAYS ARE PROBLEM SOLVERS 5.1 ISPIRATIONS

 

Inspirations and foundations for my work have been provided by:

• J. E. Benson [16] - Electrically tapered arrays with forward directivity for Sydney Opera House 1973.
• Ambrose Thompson [17] - Beam-steered O-line arrays (Martin Audio).
• Evert Start [18] - Beam-steered Intellivox arrays (Duran Audio/JBL).

 

Figure 14: Benson tapered array with rear vent from Sydney Opera House. (Metal covers removed), 12 element O-line array and 4.3 m long Intellivox array installed at Sydney’s Central Station.

 

David Gilfillan and I have worked with beam-controlled arrays since 2004 and a theme of our work with these arrays has been to design arrays that produce constant directionality from 200 Hz to 12 kHz to provide the best possible direct field coverage, along with minimal irradiation above the loudspeaker to minimise reverberation. We consider these two parameters are vital ingredients for sonic intimacy and clarity.

 

5.2 Hybrid Tapered Arrays: Australian Parliament House

 

The Australian Parliamentary Chambers presented difficult circumstances for the reinforcement of speech, including an unusually wide dynamic range, large orator-to-microphone distances, high levels of uproar noise, and architectural and acoustical constraints. These circumstances were a challenge to designing a sound system that would provide high intelligibility with faithful spectral reproduction, wide angles of coverage, and a high acoustic gain and sound pressure level capability.

 

Recognising the interdependency of all factors, a holistic process was used to design and specify hardware. The solution [19] included tapered line array loudspeakers and the use of original software.

 

High peak SPLs of 115 dBA were required to accommodate shouting matches during debate along with an equivalent acoustic distance (EAD) of 1.2 m with a mouth-to-microphone distance of 0.7 m. I believe Don Davis [20] first coined the term “equivalent acoustic distance”.

 

The system uses four bespoke hybrid arrays [21], each comprising 12 x 250 mm low/mid drivers, and 2 x 90⁰ x 40⁰ HF horns and downfill elements. Frequency tapering is applied to the low/mid drivers to minimise lobing. (see Figure 15).

 

 

Figure 15: View of House of Representatives with the four loudspeaker clusters and the arrangement of transducers in each cluster.

 

The array for the Main Table microphones uses active tapering filters entirely, while the arrays for Back and Cross Benches uses passive tapering filters. For these filters to work properly, extensive compensation of the loudspeaker impedance was necessary, which was provided by Neville Thiele.

 

Commissioning the system involved numerical optimisation of the gain structure and a novel technique using acoustic loop-gain measurements to predict equalisations that optimize the acoustic gain margin. [22].

 

5.3 Opera Theatre - Sydney Opera House

 

5.3.1 Subwoofer Array

 

In 1998, I designed a subwoofer array consisting of 6 x 480 mm JBL drivers for the Opera Theatre, which was located above the stage on the large proscenium wall.

 

To minimise the depth of the image sources on the wall, the drivers are housed in an ultra-shallow enclosure. The spacing of the drivers and the filters applied to each driver shape the beam to compensate for distance loss and minimise ceiling irradiation. Figure 16 shows the predicted vertical polar pattern of the array.

 

 

Figure 16: Predicted vertical polar response of subwoofer array.

 

 

Figure 17: Subwoofer array (right) and low/mid array (left) on proscenium wall.

 

 

Figure 18: View of subwoofer array in the factory.

 

 

Figure 19: Array of 5 x JBL 4895 LF/MF horns being installed.

 

5.3.2 Array of 5 x JBL 4895 LF/MF Horns

 

An array of five JBL 4895 comprising low and mid frequency horns was frequency-tapered to produce high directionality in the vertical plane to provide targeted sound for the dress circle in the theatre. I predicted the directionality using the published polar plot of one device as the input to my array software. Phase information was not available. The array was measured at JBL’s facility; unfortunately, the measurement process omitted the required 6 dB boost to the inner horn. Figure 20 shows the predicted and measured responses of the system. The measured and predicted polar responses are quite close without that boost.

 

Figure 20: Predicted and measured polar responses in the vertical plane of the array of 5 x 4985 horns.

 

5.4 Hybrid Beam-Steered Arrays: St Paul’s Cathedral Melbourne

 

This system uses multiple hybrid, steered-arrays with a tilted tweeter array arranged in a distributed and delayed format. Figure 21 shows a view of the cathedral.

 

As steering high-frequencies with a wide opening angle is extremely difficult with current transducers, a separate tilted tweeter array sits below the low/mid frequency array. The tweeter array consists of three dome tweeters that are electrically tapered.

 

This system is a vertical implementation of loudspeakers described by Gilfillan et al [23].

 

All-pass filters are used to decouple adjacent arrays on columns to reduce degradation of the direct-field frequency response due to phase interference.

 

We had a few anxious moments when we measured the frequency response of the first array that we commissioned.

 

The degradation of the frequency response due to reflections and diffraction of the direct field sound from the fluted columns was much worse than anticipated, and the subjective sound quality was substantially damaged.

 

We were able to significantly reduce the interference from reflections from the columns by installing terry towelling on the edges of the arrays, and we were fortunate that the towelling had such a major effect. Figure 22 shows two arrays on a column with the terry-towelling on each side of the arrays.

 

 

Figure 21: View of the cathedral nave. The arrays are the grey objects on the columns.

 

 

Figure 22: Arrays with terry towelling to reduce interference.

 

5.5 Hybrid Steered Arrays: New Zealand and Northern Territory Parliaments

 

The debating chamber area of these parliaments is covered by bespoke hybrid array clusters and are described by Gilfillan and Leembuggen [24]. Each cluster uses three separately steered arrays of drivers and tilted horn tweeters. Twenty-three amplifier channels are used per loudspeaker cluster.

 

Figure 23 shows views of the NZ parliamentary debating chamber and the Legislative Assembly of the Northern Territory Parliament and the cluster of three array .

 

Figure 24 shows the predicted vertical polar responses of the three steered arrays.

 

Figure 23: Left: view of debating chamber in the NZ Parliament; Centre: view of one of the clusters of three arrays; Right: Legislative Assembly in the Northern Territory Parliament.

 

 

Figure 24: Predicted vertical polar responses of the low, mid and upper mid-driver arrays along with the target polar response.

 

5.6 Hybrid Steered Arrays: NZ Supreme Court and Federal Court of Australia

 

In the situations of Supreme Courts of NZ and NSW and the Federal Court of Australia, our standard arrays were used, with Perspex “wings” located on each side of the array to reduce the level of sound radiated behind array towards unwanted areas. Three back-to-back arrays are employed to cover the judges and barrister tables, and all-pass filters are used to “decorrelate” these arrays by substantially increasing the density of comb-filtering nulls and peaks.

 

 

Figure 25: Supreme Court of NZ - Bespoke hybrid arrays. Ceremonial Federal Court of Australia.

 

5.7 Intellivox Arrays in Cross River Rail Project Brisbane

 

The street-level concourses of Cross-River railway station are large, cavernous spaces containing four banks of escalators. The combination of the relatively high reverberation times and large distances to listeners presented a demanding situation for the sound system to achieve an STI rating of 0.6.

 

The best solution would have been to suspend directional loudspeakers above each escalator bank with appropriate delays between the levels, but this was not deemed suitable by the project architects.

 

The solution we developed uses a pair of JBL Intellivox DSX 280HD arrays per escalator bank. Analysis and modelling of competing products in the long-steered array category showed that the DSX 280HD produced the most consistent frequency responses over the escalator area.

 

As the high-frequency beam of single DSX 280HD array is still too narrow to cover the escalator bank, two arrays per bank are used. To reduce the severity of interference between adjacent arrays, all-pass filters will be used to “decorrelate” the arrays.

 

Figure 26 shows the arrangement of two DSX 280HD arrays in a cross-sectional view of the entry cavern of Woolloongabba Station.

 

 

Figure 26: Woolloongabba Station Brisbane JBL Intellivox DSX280HD.

 

5.8 O-Lines in St Andrews Cathedral Sydney

 

The project timeline for this project was ultra-tight; only three weeks were available from the initial discussions with the Cathedral to the hosting of Royal visitors for the Easter service in 2014.

 

Given this timeline, there was no opportunity to carry out the detailed acoustic modelling that due-diligence requires, and we needed to design the system using only our experience .

 

Based on the high ratio of distance of far-to near-listeners, we selected the Martin Audio O-line arrays, arranged in a distributed and delayed format.

 

We used a combination of 6, 8 and 12 element O-line arrays with a mix of individually processed (FIR) and unprocessed arrays. As the processed arrays required individual amplifier channels, unprocessed arrays were used in less-frequented areas for cost minimisation.

 

Figure 27 shows a view of the cathedral nave.

 

 

Figure 27: View of the nave at St Andrew’s Cathedral.

 

5.9 O-Lines in Grand Concourse Sydney’s Central Station

 

The Grand Concourse at Sydney’s Central station has a mid-band reverberation time of 5 seconds and contains large areas of strongly specular surfaces.

 

The only arrangement of loudspeakers that could accommodate heritage and operational constraints involved aiming those loudspeakers at some of these surfaces. In addition, the large ratio of far-to-near listener distances required that the loudspeakers’ radiation pattern had a large opening angle.

 

Seven brands of commercial processed line-arrays were investigated using manufacturers’ software and EASE for spatial consistency of direct-field frequency response, echoes and minimising irradiation of the high roof. The combination of eight O-line curved and beam-formed arrays provided the best available solution (shown in Figure 28). Each array comprises 12 x O-line modules with beam-forming FIR processing applied to each module.

 

Figure 28: Eight O-line arrays in the Grand Concourse. With current beam steering techniques, the only way to address a requirement for a high ratio of far-near-to-far listener distances is to use a curved array.

 

Signals were fed at the 100-volt level through ultra-linear transformers, as the amplifiers were 200 m from the arrays. Identical elements in each array shared a common amplifier channel.

 

The Martin Audio D2 software allows the user to view predicted frequency responses along the vertical axis of an array. Two examples of responses are shown in Figure 29.

 

Figure 29: Examples of frequency response outputs from Martin Audio D2 modelling software.

 

5.10 Beam Formed Arrays – Cardio-Bessel Array in NSW Legislative Assembly

 

Three cardio-Bessel arrays are used to provide the primary coverage at low and mid frequencies in the New South Wales Legislative Assembly [25] . The arrays float above the debating chamber as seen in Figure 30. The Bessel array [26] of five 165 mm drivers essentially eliminates the interference effects in the long axis of the array as shown in Figure 31 . To provide the cardioid polar pattern, upper and lower drivers are arranged as shown in Figure 32.

 

During commissioning, we found that the measured acoustical separation of the upper and lower driver Bessel arrays was considerably larger than expected and too large to for the system to produce a classical cardioid/hypercardioid radiation pattern above 300 Hz. This acoustic separation problem was later explained by John Vanderkooy [27].

 

 

Figure 30: View of the NSW Legislative Assembly.

 

Using the combination of the substantial amount of digital signal processing shown in Figure 33 and the measured amplitude and phase responses of the upper and lower drivers, we were able to “rescue” the beam forming of the array and produce some 13 to 14 dB of upward attenuation above 100 Hz.

 

 

Figure 31: Frequency response of the Bessel array around the long axis.

 

 

Figure 32: Cross-sectional layout of drivers to produce the cardioid response.

 

 

Figure 33: Signal processing used to reduce rear radiation from the cardio-Bessel array.

 

Figure 34 shows the predicted responses at four angles around the array, which were subsequently confirmed by measurement.

 

Figure 34: Frequency responses of the array predicted frequency responses at different angles.

 

6 SPEECH TRANSMISSION INDEX

 

6.1 Introduction

 

The Speech Transmission Index (STI) was developed by Hermann Steeneken and Thommo Houtgast to assess the loss of speech intelligibility in a transmission channel and is well accepted. STI measures the loss of modulation within a test signal, caused by reverberation and noise.

 

Development of the STI measurement process was continued by Sander van Wijngaarden and Jan Verhave and the International Electrotechnical Commission (IEC) 60268-16 maintenance team chaired by Peter Mapp.

 

The STI has enormous strengths and a few weaknesses. An important weakness is that it is unable to properly recognise the loss of intelligibility due to poor frequency response, even when the ambient noise is low. It also has some insensitivity to echoes.

 

For readers who are not familiar with the STI concept, Figure 35 and Figure 36 provide graphical representations of the reduction in modulations.

 

 

Figure 35: Intensity envelope of speech showing fluctuations (aka modulations).

 

 

Figure 36: Reduction in modulation (reproduced from Bose documentation).

 

6.2 Word Scores and STI

 

Loss of intelligibility is important with many talkers and listeners who are non-native English speakers. The most recent correlation of word loss and STI was done by Lorenzo Morales [28] with speech degraded by reverberation. Morales’ work addressed a number of potential problems with Anderson and Kalb’s (AK) method.

 

Figure 37 shows relationships between STI value and word-scores assessed using different methods. (PB256/1000 is from the Common Intelligibility Scale.)

 

The sensitivity of STI scores to the range of word scores from 5% loss to 30 % loss can be seen by inverting Figure 37 and zooming in on part of the graph, as shown in Figure 38.

 

Table 1 lists the corresponding range in STI scores for the four relationships.

 

Morales’ relationship shows that over this range of word-score losses, the STI only changes by 0.2. Accordingly, I conclude that is possible to have reasonable STI scores, but still have considerable loss of intelligibility that is likely to be problematic for listeners.

 

Table 1: Range of STI values for word-score losses of 5% to 30%.

 

 

 

Figure 37: Relationship between STI value and word scores assessed using different methods.

 

Figure 38: Inverted/zoomed view of Figure 37

 

6.3 Effects of Degraded Frequency Responses

 

In the studies described in [29] [30], grossly shaped filters were used to degrade the quality of speech, which was broadcast in a reverberation chamber. Recordings were made in the chamber and replayed to listeners to determine the word-scores with those filters. Figure 39 shows the frequency responses of three of the filters used.

 

 

Figure 39: Frequency responses of three of the filters used to degrade the sound quality for speech.

 

The word scores were converted to STI values and compared with measured STI values with a range of possible ambient noise levels present during the word-score listening. Results are shown in Figure 40 for a possible noise level in the word-score listening situation of 36 dBA

 

Figure 40: Word scores and equivalent STI scores computed using the four relationships and the measured STI with ambient noise retrospectively added.

 

I believe the inability of STI to properly account for the effects of poor frequency response is due to:

• The low sensitivity of STI values to word score.
• An upward masking model that is much less severe than masking models such as developed by Moore and Glasberg, and Zwicker.
• There is a huge variation in the spectra of individual words in speech, which allows upward masking to be much more impactive than if speech was a constant spectrum. Figure 41 compares the spectra of five words relative to the long-term standard IEC speech spectrum.
• The use of a long-term speech spectrum, compared to a short-term spectrum, also compounds the issue. Figure 42 compares the spectra of the mean, and 10% and 90% exceedance levels of a talker within a 250 ms period with that of the long-term IEC spectrum.
• Poor frequency response in systems further reduces the relative levels between frequencies.

 

 

Figure 41: Spectra of five words relative to the long-term standard IEC speech spectrum.

 

 

Figure 42: Spectra of a talker within a 250 ms period

 

6.4 Effect of Other Masking Algorithms

 

In 2010, I investigated the effects of other masking algorithms and varying spectra of speech [31], [32]. The speech spectra of several talkers were filtered with the three above-mentioned filters and the spectra analysed in 50 ms time slices.

 

The STI scores were the computed using those 50 ms long spectra with seven different masking algorithms. Figure 43 shows the histograms of the STI scores.

 

The MPEG-1 masking models produce lower STI scores, however MPEG-1 Model 2 cannot be used as it is not level-dependent.

 

 

Figure 43: Histograms of STI scores with different spectra in 50 ms slices and different masking algorithms for the three filters.

 

6.5 STI Scores in Road Tunnels

 

The effects of poor frequency response on STI scores in the presence of high noise levels were explored in [30] using data obtained from two road tunnels. Reverberation times in the tunnels were around 8 seconds in the 250 Hz band and noise from the jet-fans ranged from 84 to 86 dBA on the roadway. A photo of the tunnel jet fan measurements is shown in Figure 44.

 

Seven filters simulating very poor frequency responses were applied to the long-term IEC standard speech spectrum. Figure 45 shows the frequency responses of those filters and inspection of them will indicate the nature of the poor sound quality that would be reproduced. The resulting speech spectra along with the tunnel noise were applied to twenty measured modulation transfer function (MTF) matrices that had been measured in the tunnels and a noise level of 85 dBA.

 

The average STI score resulting from the seven filters are listed in Table 2 along with the differences with the no-filter STI score. The STI scores degrade slightly, but the subjective intelligibility will be quite poor, especially with Filters 3 to 7.

 

 

Figure 44:

 

Table 2: Average STI scores with no filter and seven shaping filters and and fan noise of 85 dBA.

 

 

Figure 45: Filters applied to IEC long-term speech spectra.

 

7 ROOM ACOUSTICS

 

7.1 Multipurpose Auditorium in New Zealand National Library

 

An unused space on the Ground floor of the Library room was fitted out to be used for speech presentations, live music, and video. As the room was to be carpeted with plush seating, the reverberation times (RTs) at mid frequencies would be too low for classical music such as chamber ensembles. In addition, there was little opportunity to optimise room dimensions for acoustic performance.

 

Although the RTs would be low, David Griesinger’s [33] emphasis on listener engagement was forefront in my thinking. My perception of listener engagement in small rooms is that the very-early arriving sound field; in the period up to 30 ms after the first wavefront arrives, is very important.

 

Shaped reflectors on the side walls and ceiling direct early first and second order reflections towards the seating areas. Figure 46 and Figure 47 show the profiles that were used.

 

The ceiling reflectors also functioned as low-frequency Helmholtz absorbers, which were designed using our travelling wave computation method.

 

I consider that the auditorium sounds gorgeous – the sound is natural and intimate and has a sense of depth.

 

Figure 46: Plan view of room.

 

 

Figure 47: Cross-sectional view of room.

 

 

Figure 48: Views of the auditorium.

 

For interest, four of the acoustic measurements I made in the auditorium are presented below in Figure 49.

 

 

Figure 49: Selected acoustic measurements made in the auditorium, NZ National Library.

 

My comments on those measurement results are:

a) The reverberation times are relatively similar over the range 125 Hz to 8 kHz.
b) At the rear of the auditorium, the balloon burst in the 63 Hz band builds up faster than with the sound system. The reason for this is unclear, but it may be due to interference resulting from the different path lengths of the subwoofer on each side of the room to the measurement point.
c) At 125 Hz, the growth time is similar for balloon bursts and the sound system.
d) At frequencies 250 Hz and above, the growth time with the sound system is less than 20 ms to the --3dB level, whereas with the balloon source it is approximately 33 ms.
e) The growth time at the second-last row with the balloon is remarkably fast, indicating a high degree of very early sound energy.
f) As balloons cannot provide repeatable directionality performance, the results may be skewed due to these variations.
g) The C50 clarity ratios with the balloon bursts are quite high, other than at 125 Hz at the rear of the auditorium. This low ratio in the 125 Hz band is a due to reflection from the bursting balloon that occurs around 62 ms after the direct sound arrives. As this reflection does not occur in the other octave bands, its source is unknown.

 

7.2 Recording and Control Rooms at JMC Academy

 

This project involved the design of three control rooms and three live recording rooms, that were to be fitted into existing base-building shells. These rooms are used for teaching.

 

For this project, I drew inspiration and theoretical foundations from these learned people:

• Philip Newell [34] [35]: Non-environment monitoring situations and guidance for recording rooms. In 2002, I had an amazing listening experience in one of Philip’s control rooms in Oporto, Portugal.
• Bob Walker [36] [37] [38] [39]: Consideration of room modes, controlled image design and control of the very-early sound field.
• Keith Holland: Excess group delay, and modulation transfer function at low frequencies [40] [35] [41].
• Siegfried Linkwitz [42]: Modulation transfer function in listening rooms with different types of loudspeakers.
• Glyn Adams [11]: Effects of image sources in rooms as described in Section 3.6.
• Roy Alison [10]: Image sources.

 

One of the project goals was to make the control rooms non-environment spaces. Historically these types of spaces use deep sound absorption structures on the walls and ceiling. Limited space precluded use of this type of low frequency sound absorption, necessitating existing use of Helmholtz absorbers.

 

The images below in Figure 50 show aspects of various control rooms and live rooms of this project.

 

The acoustic design and performance of Control Room A is the subject of another paper I presented at RS 2022. [43]

 

 

Figure 50: Clockwise from top left: Control A, Live A, Live C, Control C, Live B, Live A

 

8 CONCLUSION

 

There have been many situations in my career in which standard design recipes or readily-available measurement or software tools have not been able to provide a suitable platform for design to proceed with integrity.

 

In those situations, I have found thinking about problems from a first-principles perspective has allowed me to develop a way forward. There is a deep well of knowledge in the fields of electroacoustics and acoustics which can provide a solid basis for a holistic design process carried out in unknown territory.

 

In that well can be found the skills and knowledge that the giants on which I stand have imparted.

 

9 REFERENCES

1. L. L. Beranek, Acoustics, American Institute of Physics, ISBN-13 978-0883184943.

2. S. Linkwitz, “Active Crossover Networks for Noncoincident Drivers,” JAES Vol. 24 Issue 1, Feb 1976.

3. S. Linkwitz, “Shaped Tone-Burst Testing,” JAES Vol. 28 Issue 4, Apr 1980.

4. S. Colam, G. Leembruggen, “A Computational Method for Analysis and Design of Acoustic Absorbers and Low Frequency Transmission Loss,” Proc. IOA Vol. 25 Part 8, 2003.

5. M. Garai, Pompoli, “A Simple Empirical Model of Polyester Fibre Materials for Acoustical Applications,” Applied Acoustics, Vol. 66 Issue 12, Dec 2005.

6. J. Vanderkooy, R. Stevens, “A Novel Single-Microphone Method of Measuring Acoustical Impedance in a Tube,” AES Convention 115, Oct 2003, Paper No. 5901.

7. G. Leembruggen, D. Gilfillan, “Comparison of Measured and Predicted Sound Absorption Properties of Polyester Fibre Insulation Using an Unusual Plane Wave Tube,” Proc. IOA Vol. 32 Part 5, 2010.

8. B. Long, R. Schwenke, P. Soper, G. Leembruggen, “Further Investigations into the Interactions Between Cinema Loudspeakers and Screens,” SMPTE Motion Imaging Journal, Nov/Dec 2012.

9. C. Hanrahan-Tan, G. Leembruggen, “Development and Use of a Low-Cost Acoustic Flow Resistance Meter,” Proc. IOA Vol. 42 Part 3, 2020.

10. R. F. Allison, “The Influence of Room Boundaries on Loudspeaker Power Output,” JAES Vol. 22, Jun 1974.

11. G. Adams, “Time Dependence of Loudspeaker Power Output in Small Rooms,” JAES Vol. 37 No. 4, Apr 1989.

12. R. H. Small, “Direct Radiator Loudspeaker System Analysis,” JAES Vol. 20 Issue 5, Jun 1972.

13. R. H. Small, “Efficiency of Direct-Radiator Loudspeaker Systems,” JAES Vol. 19 Issue 10, Nov 1971.

14. R. H. Small, “Vented-Box Loudspeaker Systems – Parts 1–4,” JAES Vol. 21 Issues 5–8, 1973.

15. G. Leembruggen, “The Design of Three Unusual Loudspeakers for the High Court of Australia,” JAES Vol. 48 Issue 5, May 2000.

16. J. E. Benson, “Theory and Applications of Electrically Tapered Electroacoustic Arrays,” IEEE Int. Conf. on Acoustics, Speech, and Signal Processing, Apr 1976.

17. A. Thompson, “Real World Line Array Optimisation,” Reproduced Sound Conference, 2006.

18. E. Start, G. van Beuningen, “Analysis of DDS-Controlled Loudspeaker Arrays by Near Field Acoustic Holography,” Reproduced Sound, 2001.

19. G. Leembruggen, D. Connor, “Design and Commissioning of Sound Reinforcement Systems for the Australian Parliament – A Holistic Approach,” JAES Vol. 44 Issue 10, Oct 1996.

20. D. Davis, C. Davis, Sound System Engineering, H.W. Sams, ISBN 10: 0672211564, 1975.

21. A. Thompson, G. Leembruggen, D. Gilfillan, “Hybrid Line Arrays – A Viable Alternative,” Proc. IOA Vol. 31 Part 4, 2009.

22. G. Leembruggen, D. Gilfillan, “A First Principles Method to Rapidly Optimise the Acoustic Gain of a Sound System with Multiple Live Microphones,” Proc. IOA Vol. 40 Part 4, 2018.

23. D. Gilfillan, M. Thompson, A. Thompson, G. Leembruggen, “Proceedings of the Extreme Steering Committee,” Proc. IOA Vol. 43 Part 3.

24. D. Gilfillan, G. Leembruggen, “Raising the Tone of the Debate: Sound Reinforcement Systems for the Northern Territory and New Zealand Parliaments,” Proc. IOA Vol. 33 Part 6.

25. G. Leembruggen, “Tailoring the Total Radiation Pattern of a Group of Loudspeakers,” Proc. IOA Vol. 23 Part 8.

26. D. B. Keele, “Effective Performance of Bessel Arrays,” JAES Vol. 38 Issue 10, Oct 1990.

27. J. Vanderkooy, “The Low-Frequency Acoustic Center: Measurement, Theory, and Application,” AES Convention 128, May 2010, Paper No. 7992.

28. L. Morales, S. Dance, B. Shield, G. Leembruggen, “Speech Transmission Index for the English Language Verified Under Reverberant Conditions with Two Binaural Listening Methods: Real-Life and Headphones,” JAES Vol. 62 Issue 7/8, Jul 2014.

29. G. Leembruggen, A. Stacey, “Should the Matrix Be Reloaded?,” Proc. IOA Vol. 25 Part 8, 2003.

30. G. Leembruggen, “Another Look at the Relationships Between Frequency Response and the Speech Transmission Index, with Respect to Word Scores and Road Tunnels,” Proc. IOA Vol. 41 Part 3, 2019.

31. G. Leembruggen, M. Hippler, P. Mapp, “Further Investigations into Improving STI’s Recognition of the Effects of Poor Frequency Response on Subjective Intelligibility,” AES Preprint 8051, 128th Convention, May 2010.

32. G. Leembruggen, M. Hippler, P. Mapp, “Exploring Ways to Improve STI’s Recognition of the Effects of Poor Spectral Balance on Subjective Intelligibility,” Proc. IOA Vol. 31 Part 4, 2009.

33. D. Griesinger, “The Importance of the Direct-to-Reverberant Ratio in the Perception of Distance, Localization, Clarity, and Envelopment,” Proc. IOA Vol. 30 Part 6, 2008.

34. P. Newell, Recording Studio Design, 4th Edition, First Published 2017, eBook ISBN 9781315675367.

35. P. Newell, K. Holland, J. Newell, “The Yamaha NS10M: Twenty Years a Reference Monitor – Why?,” Proc. IOA Vol. 23 Part 8.

36. R. Walker, “A New Approach to the Design of Control Room Acoustics for Stereophony,” AES Convention 94th, Preprint 3543, Berlin, 1993.

37. R. Walker, “Optimum Dimension Ratios for Small Rooms,” AES Convention 100, May 1996, Paper No. 419.

38. R. Walker, “Room Modes and Low-Frequency Responses in Small Enclosures,” AES Convention 100, May 1996, Paper No. 4194.

39. R. Walker, “Early Reflection Control in Stereophonic Control Room  and Listening Rooms – A Personal View,” Proc. IOA Vol. 33, 2011.

40. K. Holland, P. Newell, “Non-Minimum Phase Behaviour of Loudspeakers at Low Frequencies,” Proc. IOA Vol. 41, 2019.

41. K. Holland, P. Newell, P. Mapp, “Modulation Depth as a Measure of Loudspeaker Low Frequency Performance,” Proc. IOA Vol. 26 Part 8, 2004.

42. S. Linkwitz, “Investigation of Sound Quality Differences Between Monopolar and Dipolar Woofers in Small Rooms,” AES Convention 105, Sep 1998, Paper No. 4786.

43. G. Leembruggen, C. Hanrahan-Tan, O. Fong, “Acoustic Design of a Non-Environment Control Room for Teaching,” Proc. IOA Vol. 44 Part 3, 2022.

44. R. H. Small, “Closed-Box Loudspeaker Systems – Parts 1 and 2,” JAES Vol. 20 Issue 10; Vol. 21 Issue 1.