GB2507582A

GB2507582A - Self-calibrating audible alarm trigger

Info

Publication number: GB2507582A
Application number: GB1219936.0A
Authority: GB
Inventors: James S Swift; Nicholas Goddard
Original assignee: Stephenson Gobin Ltd
Current assignee: Stephenson Gobin Ltd
Priority date: 2012-11-06
Filing date: 2012-11-06
Publication date: 2014-05-07
Anticipated expiration: 2032-11-06
Also published as: GB2507582B; GB201219936D0

Abstract

An acoustic alarm recognition unit (1, Fig.1) comprises a microcontroller (100, Fig.1) having a sound input (110, Fig.1). In a learning mode the microcontroller processes S602 sound data representing an acoustic alarm received at the sound input to produce a test measure S604 of the acoustic alarm and to determine a test value S606 based thereon. In a live mode it processes S610 the received sound data to obtain a live measure thereof and to compare S620 the live measure with the test value to determine S630 if the acoustic alarm is sounding. Processing the sound (Fig.2) may include analogue-to-digital conversion (by ADC 116, Fig.1), band pass filtering, rectification, descriptive statistical analysis (e.g. calculating a time-varying frequency property) and may be carried out for multiple time periods (e.g. sliding windows). The unit may be used to within door control mechanisms, control panels, radio sounders or vibrating pillow alarms.

Description

An Acoustic Alarm Recognition Unit

Field of the In vent/on

The invention is in the field of alarm triggers which are able to recognise when an audible, or acoustic, alarm is activated. More specifically, the invention relates to an acoustic alarm recognition unit. The invention is useful in door control mechanisms and other devices which need to know if an audible alarm has been activated, such as control panels, radio sounders and vibrating pillow alarms. Door control mechanisms are normally operable to release a door for closure, or to close a door when an alarm, such as a fire alarm, is activated. More specifically, but not exclusively, the invention relates to stand alone units which do not have a mains power connection.

Background

US 4,520,503 was published in 1985 and discloses a tone discrimination circuit for use with audible smoke or fire detectors or similar audible devices. The tone discrimination circuit automatically emits an output electronic alarm signal, for notification of persons at remote locations (e.g. at a fire control station or fire department), upon input of a proper audio tone from the smoke detector or other audible IS device. The circuit contains in series a microphone, a two-stage audio amplifier, a frequency detector, and a tone discrimination circuit. The circuit emits an output electronic alarm signal if and only if the audio input signal has sufficient amplitude, the desired frequency, and the desired duration. In the embodiments described, the duration must be 20 seconds or longer, though the circuit allows the audio input to cease for a period of 0.25 seconds, without loss of the output signal. In this way, it may be established that the audio input signal is indeed an audio alarm emitted by the smoke or fire detector, and not other sounds which may be present in the vicinity.

EF 0635092 B1 was published as WO 93/20317 in 1993 and discloses an apparatus for retaining and releasing a closure, such as a door, and more particularly a door release arrangement on a device for holding open doors such as fire doors in the event of a fire alarm being raised. The apparatus comprises a sensor for sensing from the ambient medium an alarm signal, and is arranged to release the closure in response to the alarm signal. Specifically, there is provided means for determining whether the alarm signal (sound) is of a predetermined character, and comprises a rectifier and filter circuit and a comparator. The rectifier and filter circuit filters out any signals outside the range 500- 1000 Hz. The filtered signal is then passed to the comparator and compared with a threshold duration of 5 seconds and a threshold intensity of 65 decibels. Any signal which exceeds these thresholds causes a release arrangement to release a plunger and thus release the closure.

However, while reasonably effective in practice, the apparatus is prone to give false positives and release the closure in response to sound which is not an alarm signal.

An aim of the invention is to provide an acoustic alarm recognition unit for more accurately recognising when an audible alarm is activated, thus reducing the number of false positives and increasing the reliability of devices which use the information from the acoustic alarm recognition unit.

Summary of the In vent/on

In one aspect, there is provided an acoustic alarm recognition unit. The acoustic alarm recognition unit comprises a microcontroller having a sound input for receiving sound data. The acoustic alarm recognition unit has a learning mode and a live mode and the microcontroller is configured: (a) in the learning mode, to process sound data representing an acoustic alarm received at the sound input to produce a test measure (e.g. average sound level) of the acoustic alarm, and to determine a test value based on the test measure (e.g. a percentage level of the average sound level) then (b) in the live mode, to analyse sound data received at the sound input to obtain a live measure of the sound data and to compare the live measure with the test value to determine if the acoustic alarm is sounding.

In this way, the acoustic alarm recognition unit is self-calibrating and can more accurately recognise an acoustic alarm signal based on the actual acoustic alarm signal itself This is important as alarm sounders typically have different output characteristics, and buildings typically have different acoustic properties thereby changing the properties of the alarm signal as it propagates through a building.

Reliability is increased and the number of false positives is reduced.

Here, the term microcontroller defines a CPU, at least one type of memory and at least one peripheral circuit, such as an ND converter. The microcontroller is normally, but need not be, on a single chip.

Also, when the microcontroller decides the acoustic alarm is sounding the microcontroller may pass this information to other software in the microcontroller that then takes some appropriate action, such as allowing a door to close, closing a door, activating a sounder or deaf alarm, or passing an electrical signal to another unit, such as a control panel.

Preferably, during the learning mode the microcontroller is arranged to process the sound input by selecting band pass filter characteristics for analysing the sound data in the live mode.

Preferably, the band pass filter lower frequency is adjusted to be closer to a frequency spectrum of the acoustic alarm. Preferably, the band pass filter upper frequency is adjusted to be closer to a frequency spectrum of the acoustic alarm. In this way, more accurate recognition of an acoustic alarm is achieved. Preferably, a binary search algorithm is used to select one or both of the lower and upper frequencies. In this way, much quicker selection of appropriate band pass filter settings is achievable (for example, seconds rather than thousands of seconds if a search through 1024 filter settings were to be performed).

Preferably, the microcontroller is configured to obtain the test measure and/or the live measure by rectifying the sound data. Preferably, the rectifying occurs after filtering.

Preferably, the microcontroller is configured to obtain the test measure and/or the live measure using descriptive statistical analysis.

Preferably, the microcontroller is configured to process the sound data using one or more of the following subsets of descriptive statistical analysis: calculating an arithmetic mean of the amplitude of the sound data over a time period; calculating a time-varying amplitude of the sound data over a time period; calculating a frequency property of the sound data over a time period; calculating a time-varying frequency property over a time period. Preferably, the frequency property is a frequency spectrum.

Preferably, the microcontroller is configured to process the sound data using more than one time period to create an individual descriptive statistical analysis for each time period and optionally to calculate an overall descriptive statistical analysis using each individual descriptive analysis.

Preferably, the overall descriptive statistical analysis is calculated using a sliding window. Preferably, the corresponding result of each individual descriptive analysis is stored in a circular buffer of the microcontroller. Preferably, the descriptive statistical analysis is an average of the frequency in each time period so that frequency variation with time may be obtained.

Preferably, the acoustic alarm recognition unit comprises second memory to store the sound data over a duration of time for later analysis. Preferably, the duration of time is at least 5 seconds and up to 200 seconds. Preferably, the duration of time is between 5 and 30 seconds.

Preferably, the microcontroller comprises a user input to activate the learning mode.

Preferably, the microcontroller comprises an analogue to digital converter (ADO) to receive the analogue sound data and to convert the analogue sound data to digital sound data.

Preferably, the acoustic alarm recognition unit comprises a sound input module which includes a microphone arranged to receive acoustic sound information and to output analogue sound data to the microcontroller. Preferably, the sound input module comprises an anti-aliasing circuit. Preferably, the sound input module comprises a variable gain amplifier to adjust the level of the analogue sound data prior to the ADO.

Preferably, the microcontroller is configured, in the learning mode and prior to analysing fully the sound data, to perform a relatively quick and crude analysis on the sound data to check if the alarm is sounding.

According to another aspect, there is provided a method of recognising an acoustic alarm. The method comprising: (a) in a learning mode, processing sound data representing an acoustic alarm to produce a test measure of the sound data, and to determine a test value based on the corresponding measure. Then (b) in a live mode, processing sound data in the way determined by the learning process to obtain a live measure of the sound data and to compare the live measure with the test value to determine if the acoustic alarm is sounding.

Various optional features of the method will be appreciated from the discussion of the afore-mentioned acoustic alarm recognition unit and they are not repeated here for brevity.

Brief Descrintion of the Drawings For a better understanding of the invention, and to see how the invention may be carried out in practice, reference is now made to the following figures in which: Fig.1 is a schematic diagram of an acoustic alarm recognition unit according an exemplary embodiment; Fig.2 is a graphical representation of a process of recognising an acoustic alarm.

Fig.3 is a schematic diagram showing an input circuit of the acoustic alarm recognition unit of Fig.1 in more detail; Fig.4 is a graphical representation of a process of learning an acoustic alarm using two processes; Fig.5 is a schematic diagram showing further circuitry of the acoustic alarm recognition unit of Fig. 1 in more detail; and Figs.6 and 7 are flow charts illustrating a method of recognising an acoustic alarm.

Detailed Description

Reference is now made to the above-listed drawings to describe an exemplary acoustic alarm recognition unit and method.

In one embodiment of the invention, there is provided an acoustic alarm recognition unit 1 comprising a microcontroller 100 as shown schematically in Figs.1-3. The microcontroller 100 comprises a CPU 112, a memory 114, an analogue to digital converter (ADC) 116 and a bus 118 interconnecting the aforesaid components on the microcontroller 100. The microcontroller 100 comprises various inputs and outputs (not all of which are shown), including an analogue input 110 for receiving sound data.

The sound data is received at the input 110 as an analogue signal and the ADO 116 is used to convert the analogue signal to a digital signal for processing by the CPU 112 and memory 114. Fig.2 is a graphical representation of a process of recognising an acoustic alarm.

The acoustic alarm recognition unit 1 has a learning mode and a live mode. The microcontroller 100 is configured in the learning mode to first store and then process sound data representing an acoustic alarm received at the input 110 to produce a first or test measure of the acoustic alarm, and to determine a test value based on the corresponding test measure. The microcontroller 100 is configured in the live mode to process sound data received at the sound input 110 in the same or very similar way to obtain a live measure of the sound data and to compare the live measure with the test value to determine if the acoustic alarm is sounding.

In this way, the acoustic alarm recognition unit 1 is self-calibrating and can more accurately recognise an alarm signal based on the actual alarm signal itself. This is important as alarm sounders typically have different output characteristics, and buildings typically have different acoustic properties thereby changing the properties of the alarm signal as it propagates through a building. Reliability is increased and the number of false positives is reduced.

For example, the process may use a relatively wide band pass filter (say for example 200 Hz to 4kHz) and then take an arithmetic mean of the amplitude over a set time period. In this example, the test value is a threshold of 70% of the arithmetic mean, and the set time period is 10 seconds, but other thresholds and periods are possible.

The microcontroller 100 also comprises an output 120 for outputting control data indicative of whether the acoustic alarm is sounding, as will be described further later.

Fig.3 is a schematic diagram showing the acoustic alarm recognition unit 1 in more detail.

Specifically, the acoustic alarm recognition unit 1 comprises a sound input module 200 arranged to receive acoustic sound information and to provide analogue sound data to the analogue input 110 of the microcontroller 100.

For this purpose, the sound input module 200 comprises a microphone 210, a pre-amplifier 220, a programmable/switchable/fixed gain amplifier 230 and an anti-aliasing filter 240 having a fixed gain.

The microphone 210 transforms incident sound waves into a corresponding analogue electrical signal, herein called analogue sound data. The pre-amplifier 220, programmable/switchable/fixed gain amplifier 230 and anti-aliasing filter 240 are low-power devices and are powered by a 1.8 volt regulator (not shown). Mid-point references of 0.9 volts are provided by amplifiers not shown, to the amplifiers 220, 230 to remove the need for low-impedance voltage dividers which would consume more power. The microcontroller 100 is able to shut down the analogue input circuit 200 by outputting a regulator enable signal via control line 202 to turn off the 1.8 volt regulator (not shown). In this way, power can be saved during operation, prolonging battery life. This is an important practical consideration for door holders and other related products.

The analogue input circuit is designed to amplify the sound data to a level suitable for input into the ADO 116 of the microcontroller 100. The anti-aliasing filter 240 is designed to prevent noise above half the sampling rate of the ADC from causing aliasing. In the example, the ADO is configured to sample at 20 kHz. For maximum performance, the amplitude of the signal going into the ADO 116 should be close to its maximum for the maximum sound level. To this end, the programmable/switchable/fixed gain amplifier 230 is a programmable amplifier and is controlled via a serial peripheral interface (SPI) interface 232 from the microcontroller 100. Alternatively, the programmable/switchable/fixed gain amplifier 230 may be switch selectable or of a fixed gain depending on cost and implementation considerations.

The microcontroller 100 is able to measure the output of the pre-amplifier 220, which is DO coupled, via connection 222 between the output of the pre-amplifier 220 and the microcontroller 100. Hence the microcontroller 100 may be configured to detect if the microphone 210 and pre-amplifier 220 are operational.

Once the analogue sound data is digitised by the ADO 116, the microcontroller 100 is configured to process the digital sound data output by the ADO 116 by filtering the digital sound data using a configurable band pass filter implemented by the microcontroller using digital signal processing (DSP) techniques. In this example, the band pass filter is 8 pole and has a sharp cut-off. The microcontroller 100 has a library of suitable DSP functions for this purpose, and the microcontroller is configured to process the digital sound data by rectifying the sound data so that meaningful average values can be measured.

Fig.4 is a graphical representation of a process of learning an acoustic alarm signature using two processes. However, in practice, many such processes may be used to give added reliability.

In this embodiment, the microcontroller 100 is arranged to process the digital sound data at least twice using first and second processes (Process #1, Process #2) to obtain first and second measures (Measure #1, Measure #2) of the sound data. The microcontroller 100 is arranged to choose which of the at least first and second measures is the most suitable measure for use in later identifying the acoustic alarm. In this way, depending on the properties of the acoustic alarm and the acoustic properties of an installation, a most suitable measure can be used thereby improving reliability. Here, the term most suitable means least likely to give false alarms, i.e. most tuned to the unique characteristics of a particular alarm sounder.

In Fig.4, Process #1 uses a band pass filter having a band including higher frequencies than the band pass filter used in Process #2. As a result, in this example, the average (Average #1), or more particularly arithmetic mean, of Process #1 is higher than the arithmetic mean of Process #2 (Average #2). Likewise, the frequency (f 1) or mean frequency resulting from Process #1 is higher than the frequency (f2) or mean frequency resulting from Process #2. The microcontroller 100 is configured to choose which process is the most suitable based on the resulting measurements of the acoustic alarm (Average #1, Average #2, f 1, f2) or the process variables (i.e. the band pass filter settings). For example, the highest average frequency or highest frequency may be used. A weighting may be used to weight one of the measurements. The weighting may be based on one of the other measurements, or on parameters used in the process, or on the process itself where, for example, the arithmetic mean is replaced by a time-varying amplitude analysis approach. The weighting may be based on the ability of the process and parameter settings to discriminate against common ambient noises sources -such as vacuum cleaners, music and voices -as determined in separate tests.

It follows that in the live mode, the microcontroller 100 is arranged to process the sound data using the most suitable process. There is also a default process that will be used should a user not use the learning facility. In the example embodiment there are a finite number of ways of processing the sound data, and the microcontroller 100 is configured to apply a weighting to each process and the weighting may be based on parameters used in the measure. Typical parameters used in the measure would include bandwidth of the band pass filter, and start and corresponding end frequencies. Also, the weighting may be based on the type of descriptive statistical analysis used.

In one example embodiment, the recorded sound data is first filtered using a pass band of 200 Hz - 4,000 Hz, and the sound level is then computed by rectifying' and averaging. The lower frequency is then varied from 200 Hz to 200 Hz + 3,200 Hz using a binary search of 5 steps. If the frequency can be increased by the step without reducing the average sound level by more than a desired level, say 5%, the step is added to the base frequency. If the average is reduced by more than the 5%, the step is not added. Given the range of 3,200 Hz to be searched, the step sizes are 1600, 800, 400, 200 and 100. The algorithm is therefore: Top 4000 Base = 200 Step = 1600 CurrentAverage = SoundLeve]. (Base, Top) For I = 1 to 5 If SoundLevelL(Base + Step,Top) >= X*CurrentAverage CurrentAverage = SoundLevel (Base+Step, Top) Base Base + Step Step = Step / 2 LowerFrequenoy Base Because the aim is to maximise the lower frequency of the filter, starting at a lower frequency, weighting is automatically given to higher frequency settings. Therefore, in this way, a number of processes are performed on the sound data and the most suitable process is chosen.

A similar method is used to minimise the upper frequency of the filter.

The 11 evaluations required by this algorithm give the same frequency resolution as searching through 1024 filter settings, which would take an unacceptably long time (approximately 5000 seconds).

The trigger and sleep levels are now set as a percentage of the average sound level when processed using these filter settings in the live mode.

The above values are typical and may be varied.

To explain further the concept of processing the sound data twice, differently, either of the following is meant: using at least two different methods, or ways of deriving a test measure of an alarm signal; and using the same method but with at least two different sets of parameters (for example bandpass filter settings such as lower frequency, upper frequency, etc.). The microcontroller 100 may be configured to analyse the sound data using descriptive statistical analysis and in some configurations is configured to use two different types of descriptive statistical analysis (in other words, using two different methods or ways). For example, the microcontroller 100 is configured to analyse the sound data using one or more of the following subsets of descriptive statistical analysis: calculating an arithmetic mean of the amplitude of a rectified version of the sound data over a time period; calculating a time-varying amplitude of a rectified version of the sound data over a time period; calculating a frequency property of the sound data over a time period; calculating a time-varying frequency property over a time period. The frequency property may be a frequency spectrum.

Thresholds may be used to judge arithmetic means. Additionally or alternatively, the microcontroller may vary the parameters of the band pass filter as described above.

In this way, the acoustic alarm recognition unit 1 may be able to teach itself the most suitable way of detecting an acoustic alarm. For example, acoustic alarm recognition units I may be supplied with default settings that include wide frequency settings and low thresholds. The acoustic alarm is activated and the unit 1 instructed to learn via the learning mode. In one example, the acoustic alarm recognition unit 1 then processes the sound data using a range of filter settings and considers the averages for each filter. Using a performance criteria which prioritises higher minimum frequency of the sound data, the most suitable filter setting is selected. A higher minimum frequency is preferred, as most noise sources are at a lower frequency than typical alarms. A suitable threshold for later use in the live mode is then calculated, as a percentage of the arithmetic mean achieved (i.e. the test measure). In this example embodiment, the microcontroller 100 would typically sample for 10 seconds and set a threshold value of 70 per cent of the test measure, but these parameters are of course configurable. The sampling time may be from 2 to 60 seconds and the threshold may be 40 to %.

Also, the microcontroller 100 is configured to process the sound data using more than one time period to create individual descriptive analysis for each time period and to calculate an overall descriptive statistical analysis using the individual descriptive analysis for each time period. In this embodiment, the overall descriptive statistical analysis is calculated using a sliding window and a circular buffer.

These features add further parameters such as the number and time of each time period, the width of the sliding window and the length of the circular buffer, and which may be used in the learning mode to determine a or the most suitable measure for the live mode. Typically a one second period with a buffer of 10 giving a 10 second window is useful.

It may be that variation of frequency with time is also selected to be measured. In this case, the frequency time variation of the sample during learning is stored. The frequency is averaged over short time periods, such that the period length is the learning sample length divided by N, where N is in the range 20 to several 100. N must not be too large, otherwise the computational load later in the process will be too high. Equally, the time periods should not to so short that the frequency measurement suffers excessive noise degradation. The average value of this sequence is obtained.

The RMS value of the sequence about its mean is obtained.

In live mode, a frequency time sequence is recorded, using the same length of time interval as in learning mode but in one example with half the number of points. The average of this sequence is obtained, and the maximum value of the cross correlation function from the learnt and live sequences about their mean is computed. The closeness of the RMS value compared to the maximum cross correlation value indicates if the live frequency time variation matches that of the learning sequence.

This is described mathematically as follows: Learning frequency series L[i], i = ito N L[i] AvL = (Equation 1)

N

RMS = +X(L[i]_AvL)2 (Equation 2) Live frequency series R[i], i = ito N/2 R[i] AvR = (Equation 3) N /2 Max cross correlation = I I i-X/2 N!2 maxL (L[i + k] -AVLXR[i] -A vR),J (Equation 4) 1V/2 1=1 The live mode sequence R will operate as a sliding window. Conceptually, when the N12 points are obtained, the maximum correlation will be computed. If the value does not indicate a match, then the first point of the R sequence will be discarded, the other points shifted down one, and the last point replaced by a new value, and the test repeated as required.

Fig.5 is a schematic diagram showing further circuitry of the acoustic alarm recognition unit 1 of Fig. 1.

The acoustic alarm recognition unit 1 comprises second memory 310 to store the sound data over a duration of time for subsequent or later analysis. In this embodiment, the duration of time is up to 200 seconds and is preferably between 5 and 30 seconds. Of course, the duration of time will vary depending on the capacity of the memory 310 and the sampling rate of the ADC 116. The memory 310 is a Flash memory having a capacity of 8 MB, and the memory 310 is connected to the microcontroller 100 via a serial peripheral interface (SPI) 312. The acoustic alarm recognition unit 1 comprises a user input 320 to activate the learning mode. In this embodiment, the user input 320 is a a push button to set the learning mode with a time out to return to the live mode. However, it is anticipated that a remote signal can be used to activate the learning mode, for a period of time. It is anticipated that a fire control panel (as discussed later) would control whether the acoustic alarm recognition unit 1 is put into the learning mode. A sliding switch could also be used.

Also, the microcontroller 100 is configured, in the learning mode and prior to analysing fully the sound data, to perform a quick analysis on the sound data to check if the alarm is sounding using a crude measure.

The acoustic alarm recognition unit 1 may be battery powered or mains powered. When battery powered, two alkaline or other suitable type of batteries are used to provide a 2 to 3.6 volt supply. A suitable converter is used when connected to the mains electricity supply. A DC supply 332 (or an AC supply) may be provided instead of or in addition to the battery 330.

Battery Life Maximisation Prolonging battery life is a very important consideration, and if in the live mode a continual checking were to be performed, then the acoustic alarm recognition unit 1 may have an unacceptably short battery life. To achieve both the required battery life and accuracy and reliability, the acoustic alarm recognition unit 1 sleeps for a number of seconds. The acoustic alarm recognition unit 1 then wakes up and checks the sound input for a very short time period of a fraction of a second. If the average of the sound level is less than or equal to a quiet threshold, then the acoustic alarm recognition unit 1 repeats the sleep and listen cycle. If the sound level is above the quiet threshold then the unit stays awake and performs the full measure as described above.

When the acoustic alarm signal is no longer detected, then the acoustic alarm recognition unit 1 repeats the sleep and listen cycle.

Typically, a sleep period would be 20 seconds and a short-listen period would be 1/20 of a second.

Stand-Alone Acoustic Door Holder The acoustic alarm recognition unit 1 may be implemented within a door holder (not shown). In this example, the door holder uses a magnet motor drive system 340 shown in Fig.5. One example of a suitable, but not essential, magnet motor drive system is shown in patent application number PCT/G82005/003893 and the teaching related to the magnet motor drive system is incorporated herein by reference.

When the acoustic alarm recognition unit 1 recognises an acoustic alarm, the microcontroller 100 is arranged to output a suitable control signal via the control output 120 to cause the door to be released.

After a set time, or by a user pressing a door release button, the above-mentioned measure is performed to determine if there is an acoustic alarm present. If the measure is negative, then the door holder is reactivated, and the acoustic alarm recognition unit 1 repeats the sleep and listen cycle.

Acoustic Trigger The acoustic alarm recognition unit 1 may be implemented as a fire input to a Salamander (Registered Trade Mark) control panel (not shown) but any control panel or safety device would appear suitable if appropriately modified or as necessary. In this implementation, the acoustic alarm recognition unit 1 has a 2-way communication port 350 shown in Fig.5. An output indicative that an acoustic alarm has been recognised is then output over the port 350.

The panel may be powered from a mains supply, and in that case the panel may communicate to the acoustic alarm recognition unit 1 that sleep mode is not required, and the full alarm recognition test is run continuously. The panel may indicate that the mains supply has failed, and may communicate to the acoustic alarm recognition unit 1 that sleep mode is required. The acoustic alarm recognition unit I then uses the sleep listen cycle as previously described.

Fig.6 is a flow chart showing a method of recognising an acoustic alarm.

Here, the method at step 600 first assesses whether the acoustic alarm recognition unit 1 is in a learning mode or in a live mode. If during the determination, the acoustic alarm recognition unit 1 is in the learning mode, then the method moves to step 602. At step 602 the acoustic alarm recognition unit 1 processes sound data representing an acoustic alarm. For example, using the binary search algorithm already described. At step 604 the acoustic alarm recognition unit 1 obtains at least one measure of the sound data as a result of the processing at step 602. At step 606 the acoustic alarm recognition unit 1 calculates a test value based on the corresponding measure.

Once the test value has been calculated at step 606, the acoustic alarm recognition unit 1 once again checks to see whether it is in the learning mode or in the live mode. If, during this determination at step 600, it is determined that the acoustic alarm recognition unit 1 is in the live mode, then the method proceeds to step 610. At step 610 the acoustic alarm recognition unit I processes sound data using the same process or similar as that obtained during the learning mode. Then, at step 620, the acoustic alarm recognition unit 1 compares a result of the processing at step 610 with the test value previously discovered in the learning mode. It is then determined, at step 630, whether or not an alarm has been detected. If not, the method returns to step 600 once again. If an alarm has been detected at step 630, then the method proceeds to take action at step 640, before returning once more to step 600 and determining whether or not the acoustic alarm recognition unit is in the learning mode or not. Optionally, the sleep and listen cycle is used as described before, and is shown in Fig.7.

As the skilled reader will appreciate, further details of the method are derivable from the previous discussion of the acoustic alarm recognition unit 1 itself.

Fig.7 is a flow chart illustrating a sleep/wake cycle of the acoustic alarm recognition unit 1.

Firstly, at step 700, the acoustic alarm recognition unit I is in a sleep mode. Here, power to the sound input module 200 is shut-off and the microcontroller 100 enters into a low power state. Then, at step 710, the acoustic alarm recognition unit 1 waits for a sleep period, in this example approximately 20 seconds. However, the sleep period may vary between a fraction of a second and one minute. Ideally, the sleep period is between 5 and 20 seconds. Once the sleep period has elapsed, as counted by the microcontroller 100, the acoustic alarm recognition unit 1 performs a quick test on the sound data present at the microphone 210. This occurs at step 720. Then, at step 730 it is determined whether or not the sound data is above the threshold. This threshold could be the minimum frequency or a minimum average amplitude measured for a fraction of a second. Here, it is preferable that the test is conducted for 1/201h second, but any other suitable value could be used. If the test reveals that the sound data is below or equal to this threshold, then the acoustic alarm recognition unit 1 re-enters the sleep mode at step 700. If the sound data is above this threshold, then the acoustic alarm recognition unit 1 enters the live move at step 740. The live mode has already been discussed with reference to Figs.2 and 6, any skilled person would realise that the acoustic alarm recognition unit 1 could be switched into the learning mode at any time.

In fact, the precise order in which all of these steps can be carried out could be altered as would be apparent to the skilled person.

Additional Aspects of the Learning Mode The alarm recording for analysis in learning mode can be started in two ways: a) Option A is designed to be useful when a number of door holders are to be taught at the same time. On the assumption that the alarm is not sounding when the unit's learning mode is activated, the amplifier gain is selected such that the background sound level only uses a small percentage of the dynamic range of the circuit. The background sound level is then measured, and the unit waits until the alarm is activated, indicated by a significant increase in the sound level.

b) The alternative method, option B, simply does the above when the learn mode is activated by a switch or other control signal. For example, a user would press a button and after a delay (preset) of sayS seconds, the analogue sound input (including the alarm sound) is sampled and stored for analysis as in option A. For both of these options, a time delay may be inserted before recording starts. This time delay is to allow the cover of the unit to be refitted, and the operator to remove themselves from the vicinity of the unit.

The next stage in the process is that once an alarm is sounding, the amplifier gain is adjusted such that the alarm signal uses 80% of the dynamic range of the AiD convertor. When this gain has been set, the alarm sound is recorded for a specified period (e.g. 10-30 seconds).

The next stage is to identify the frequency range of the alarm signal, with the aim of making the range as small as possible to minimise the possibility of false triggers. To avoid the need to store a large number of digital filter parameters, the software now computes the filter coefficients and integer parameters from the required pass band frequencies. Because of the large number of possible pass band frequencies, the software uses a binary search algorithm as mentioned before to select first the lower filter frequency and then the upper one.

Although the invention has been described above with reference to one or more preferred embodiments, it will be appreciated that various changes or modifications may be made without departing from the scope of the invention as defined in the appended claims.

Claims

Claims: 1. An acoustic alarm recognition unit comprising: a microcontroller having a sound input for receiving sound data; wherein the acoustic alarm recognition unit has a learning mode and a live mode and the microcontroller is configured: (a) in the learning mode, to process sound data representing an acoustic alarm received at the sound input to produce a test measure of the acoustic alarm, and to determine a test value based on the test measure; and (b) in the live mode, to analyse sound data received at the sound input to obtain a live measure of the sound data and to compare the live measure with the test value to determine if the acoustic alarm is sounding.
2. The acoustic alarm recognition unit of claim 1, wherein during the learning mode the microcontroller is arranged to process the sound input by selecting band pass filter characteristics for analysing the sound data in the live mode.
3. The acoustic alarm recognition unit of claim 2, wherein the band pass filter lower frequency is adjusted to be closer to a frequency spectrum of the acoustic alarm.
4. The acoustic alarm recognition unit of claim 2 or claim 3, wherein the band pass filter upper frequency is adjusted to be closer to a frequency spectrum of the acoustic alarm.
5. The acoustic alarm recognition unit of any preceding claim, wherein a binary search algorithm is used to select one or both of the lower and upper frequencies of the band pass filter.
6. The acoustic alarm recognition unit of any preceding claim, wherein the microcontroller is configured to process the sound data by first rectifying the sound data.
7. The acoustic alarm recognition unit of any preceding claim, wherein the microcontroller is configured to process the sound data using descriptive statistical analysis.
8. The acoustic alarm recognition unit of any preceding claim, wherein the microcontroller is configured to process the sound data using one or more of the following subsets of descriptive statistical analysis: calculating an arithmetic mean of the amplitude of a rectified version of the sound data over a time period; calculating a time-varying amplitude of a rectified version of the sound data over a time period; calculating a frequency property of the sound data over a time period; calculating a time-varying frequency property over a time period.
9. The acoustic alarm recognition unit of any claim 8, wherein the frequency property is a frequency spectrum.
10. The acoustic alarm recognition unit of any preceding claim, wherein the microcontroller is configured to process the sound data using more than one time period to create individual descriptive analysis for each time period and to calculate an overall descriptive statistical analysis using each individual descriptive analysis.
11. The acoustic alarm recognition unit of claim 10, wherein the overall descriptive statistical analysis is calculated using a sliding window.
12. The acoustic alarm recognition unit of claim 11, wherein the corresponding result of each individual descriptive analysis is stored in a circular buffer of the microcontroller.
13. The acoustic alarm recognition unit of any preceding claim, wherein the acoustic alarm recognition unit comprises second memory to store the sound data over a duration of time for later analysis.
14. The acoustic alarm recognition unit of claim 13, wherein the duration of time is at least 5 seconds and up to 200 seconds.
15. The acoustic alarm recognition unit of claim 14, wherein the duration of time is between Sand seconds.
16. The acoustic alarm recognition unit of any preceding claim, wherein the microcontroller comprises a user input to activate the learning mode.
17. The acoustic alarm recognition unit of any preceding claim, wherein the microcontroller comprises an analogue to digital converter (ADO) to receive analogue sound data and to convert the analogue sound data to digital sound data.
18. The acoustic alarm recognition unit of any preceding claim, wherein the acoustic alarm recognition unit comprises a sound input module which includes a microphone arranged to receive acoustic sound information and to output analogue sound data to the microcontroller.
19. The acoustic alarm recognition unit of claim 18, wherein the sound input module comprises an anti-aliasing circuit.
20. The acoustic alarm recognition unit of claim 18 or 19, wherein the sound input module comprises a variable gain amplifier to adjust the level of the analogue sound data prior to the ADO.
21. The acoustic alarm recognition unit of claim 18, 19 or 20, wherein the microcontroller is configured, in the learning mode and prior to analysing fully the sound data, to perform a relatively quick and crude analysis on the sound data to check if the alarm is sounding.
22. A method of recognising an acoustic alarm, the method comprising: (a) in a learning mode, processing sound data representing an acoustic alarm to determine a test value based on the corresponding measure; and (b) in a live mode, processing sound data to obtain a live measure of the sound data and to compare the live measure with the test value to determine if the acoustic alarm is sounding.
23. A computer-readable recording medium having recorded thereon instructions which would cause a processor to execute the method of claim 22.