Speech Intelligibility - GOLD LINE

1 Situations where egress is complexor difficult, such as in high-risebuildings or large factories, humanvoice is often used to provide informa-tion. Failure to understand messagecontent can result in several ways. Amessage that is not intelligentmay notbe understood. A message spoken inSpanish to an audience that only under-stands Cantonese will not be under-stood. A person talking rapidly or witha Speech impediment can cause a mes-sage to not be understood. Even a well-spoken, intelligent message in the language native to the listener can bemisunderstood if it is not audible or ifits delivery to the listener is last failure mechanisms are thebasis for the specification, modeling,and measurement of Speech PROBLEMF ires such as the King s Cross fire inLondon in 1987 and an apartment fire inYork, Ontario, have been cited as situa-tions where the lack of intelligible voicecommunication to occupants was a con-tributing factor in the , 2We oftensee paging systems in places such as air-ports and meeting spaces with speakersevery eight to twelve feet (three to fourmeters).

2 How will the Speech intelligibil-ity of the adjacent fire alarm system com-pare when it has speakers spaced 40 to70 feet (10 to 20 meters) apart?No one argues that a tone signal mustbe audible to the listener and that a voicetransmission must be intelligible. Dis-agreements regarding audibility led thefire alarm industry to adopt audio indus-try definitions and measurement moved the industry from using sub-jectiveevaluations of audibility to objec-tivemethods. In 1997, the Notification AppliancesCommittee of NFPA 72 began workingwith the audio industry to learn moreabout Speech Intelligibility and how toestablish objective performance require-ments for emergency voice alarm com-munication (EVAC) systems. The goalwas to define Speech Intelligibility performance in a way that could be objectively measured, eliminating sub-jective IS Speech Intelligibility ?

3 Figure 13is useful in understanding thepath of a voice signal from a talker to alistener. The figure shows the types of errorthat can be introduced into the messageat each stage. Problems or faults have acumulative effect on message under-standing. For example, a person mightspeak with an accent but still be under-stood by a listener who is face-to-facewith the talker. The communications system might add distortion that results inthe message not being understood. Orperhaps it s understood when there is lit-tle or no background noise, but not un-derstood when there is background are addressing the twoends of the communications chainshown in Figure , 5 For the purposes ofthis article, speed of talking, language,and talker articulation are not directly addressed. They are indirectly addressedbecause a system that reliably delivers amessage, with a limited amount of distor-tion, reverberation, and echo, is morelikely to be understood even when atalker introduces problems or when a listener has impaired hearing.

4 A systemwith a higher degree of Intelligibility canoffset some, but not all, deficiencies introduced by the talker or the Intelligibility is the measure ofthe effectiveness of Speech . The mea-surement is usually expressed as a per-centage of a message that is Intelligibility does notimply Speech quality. A synthesizedvoice message may be completely under-stood by the listener, but maybe judgedto be harsh, unnatural, and of low qual-ity. A message that lacks quality may stillbe AFFECTING SPEECHINTELLIGIBILITYFor Speech to be intelligible, it musthave adequate audibility(sound pres-sure level) and adequate clarity. For audibility, we are concerned withthe signal-to-noise ratio. voice is highlymodulated, and so while intelligibilitymeasurements do incorporate audibility,it is not to the same standards used foraudibility of tone generating IntelligibilityTalkerMicrophoneMixerAmpl ifierRoomAmplifierListenerAssumednormalA ssumednormalIntelligibility measuresLanguagespeedarticulationNoise reverberation echoesLanguagehearingBandwidth distortionFigure 1.

5 voice Signal Path (Courtesy of K. Jacob, Bose Professional Systems3)17 Fire Protection EngineeringNUMBER16 Thus, a tone and a voice message that areboth perceived as equally loud may haveconsiderably different readings on a dBor dBA meter using fast or even slow timeconstants. That is one reason that audibil-ity measurements are not required by theNational Fire Alarm Codefor voice are the smallest phoneticunit capable of conveying a distinction inmeaning in a particular language and areinstrumental in accurate word are the mof matand thebof batin English. Clarity is the propertyof sound that allows phonemes to be dis-tinguished by a is thefreedom of these sound units from distor-tion introduced by any part of the soundsystem or environment. Recently, a cellular telephone company has im-plemented a television ad campaign play-fully pointing out the very real problemof phoneme clarity.

6 Clarity can be reduced by: 1) amplitudedistortion caused by the electronics/hard-ware; 2) frequency distortion caused byeither the electronics/hardware or theacoustic environment; and 3) time domaindistortion due to reflection and reverbera-tion in the acoustic environment. Designers and engineers have thegreatest effect on Speech Intelligibility bytheir choice of equipment, the numberand distribution of loudspeakers, and thepower at which they are Speech INTELLIGIBILITYThe system hardware and the acousticenvironment cannot be separated whenevaluating Speech Intelligibility . Installa-tion choices, such as wire size and rout-ing, affect power levels and inducednoise. Mounting locations and surfaces affect sound fields, and construction materials and furnishings affect acousticparameters. Thus, the performance metricfor Speech Intelligibility must assess all ofthe requisite Electrotechnical Commis-sion (IEC) and International Standards Or-ganization (ISO) standards already incor-porate objective methods for evaluatingspeech Intelligibility .

7 The standard, IEC60849, Sound systems for emergency pur-poses,8is similar to NFPA 72. Some of themethods recognized in the standard aresubject-based, and others use instrumen-tation. ISO 9921 also references estab-lished each of the recog-nized methods, there already exists aninternationally accepted standard for thetest IEC standard includes a chart thatequates the scales for each of the differ-ent test methods to a common scalecalled the Common Intelligibility Scale(CIS). Evaluation of Speech intelligibilitymay use any one of several methods citedin the standard. Four of these methodsuse test instruments. Three subject-basedmethods are also permitted. One methodhas both a subject-based solution and aninstrument-based solution. These aresummarized below in Table 1. For the four instrument-based solutions,at present there are at least six differentinstruments available from four differentmanufacturers.

8 Consult the references formore detail on each of the test recommended minimum perfor-mance level for EVAC systems is that theaverage CIS score, less one standard devia-tion, be or greater. This permits devia-tions, does not require an exact score, andMethod STI Speech Transmission IndexRASTI Rapid Acoustics Speech Transmission IndexPB Phonetically Balanced WordScoresMRT Modified Rhyme TestAI Articulation Index%ALcons Articulation Loss of ConsonantsStandard Ref. in IEC 60849 IEC 60268-16 The objective rating of Speech Intelligibility byspeech transmission index, 1998 IEC 60268-16 The objective rating of Speech Intelligibility by Speech transmission index, 1998 ISO/TR 4870 Acoustics The construction and calibration ofspeech Intelligibility tests, 1991No reference givenANSI S , Methods for the calculation of thearticulation index, 1969 ANSI S , Methods for the calculation of thespeech Intelligibility index (SII), 1997 Peutz, , Articulation loss of consonantsas a criteria for Speech transmission in a room, J.

9 Aud. Eng. , 12, December 1971 CommentsThis is an objective, instrument-based hardware and software for measurement and in a computer-based solution, as a feature of some multi-function audio analysis equipment, and as a handheld is an objective, instrument-based STI in a handheld is an objective, subject-based Method for measuring the Intelligibility of Speech overcommunication systems, 1989, is a better reference for evalua-tions using the English language. Notification Appliances Chapterpermits ANSI use, although ISO/TR 4870 is also is an objective, subject-based standard listed. ANSI notes that the method has the same limits as given in ISO/TR 4870 (PB).Good reference is ANSI Method for measuring the Intelligibility of Speech over communication systems, is an objective, instrument-based 1969 version is referenced. This has been updated to the 1997 hardware and software for measurement and is an objective, instrument-based method or an objective,subject-based in a computer-based 1.

10 Speech Intelligibility Test that approximately 84% of thespace has a score of or better as-suming a normal distribution of the 2002 edition of the NFPA 72 Hand-bookcontains a discussion of why a CIS was used as a , DESIGN, INSTALLING,TESTING, AND USINGA reliable communication system mustbe properly planned, designed, and in-stalled. Testing uncovers faults and allowscorrections to be made, but also showssuccessful techniques for future issue that designers and authoritiesmust face when planning a system is thequestion of where intelligible voice com-munication is needed. In a large space used for public meet-ings, conventions, and trade shows, anEVAC system needs to be reliably intelligi-ble because it is intended to give informa-tion to the general public that is not famil-iar with the space. In large public spaces,a person should not have to move anygreat distance to find a place where theycan understand the message.