|Can You Hear (and understand) Me Now?
10 key issues affecting the intelligibility of voice communications
NFPA Journal®, November/December 2010
By Robert P. Schifiliti, P.E., FSPE
Fire alarm systems that use voice to inform occupants and direct their movements have been a staple of fire protection for decades. In recent years, though, as so-called all-hazards communications systems—those designed not just for fires, but for any kind of emergency situation—have proliferated, a wealth of questions has arisen about how to effectively integrate voice into those systems. The changes have come rapidly and have affected disciplines and industries that previously had little knowledge, or need, for such systems. Significant changes have also been made to the code—NFPA 72®, National Fire Alarm and Signaling Code®—that covers these systems. It’s no wonder that planners, designers, authorities, installers, and users are struggling to understand and apply all of this new information.
ECS, TWO WAYS
NFPA 72, National Fire Alarm and Signaling Code, broadly divides emergency communications systems (ECS) into one-way and two-way systems. One-way systems include traditional fire alarm voice systems as well as systems used for other hazards. One-way systems are also divided into those within a building and those that either broadcast voice messages outside to a wide area or those that send messages to specific recipients, typically using text messaging, email, or mass dialing and delivery of recorded voice messages.
Two-way systems include traditional firefighter/emergency forces telephones as well as systems that enhance the use of emergency forces radios within a building or area. The code has also expanded to include requirements for two-way communications systems serving elevators and areas of refuge that have been required by building codes for some time, but have not been covered by any performance or installation standards prior to the 2010 edition of NFPA 72.
The confusion can start with even the most basic terminology. "Emergency communications system" (ECS) and "mass notification system" (MNS) are often used interchangeably, but they are not the same thing. The 2010 edition of NFPA 72 created the new, broader category of ECS to include MNS and a variety of other emergency systems. NFPA 72 defines an ECS as "a system for the protection of life by indicating the existence of an emergency situation and communicating information necessary to facilitate an appropriate response and action," and most forms of ECS rely on the use of voice as their primary messaging and communication strategy.
The needs of a growing array of ECS users, from the military to college campuses to workplaces, are forcing many of us to reconsider how voice communication is used in these systems. The problem is that many voice communications systems are still designed around the principles of audibility rather than intelligibility—another point of confusion. Audibility means you can hear something, such as a fire alarm. Intelligibility means you not just hear it, but understand it. It’s the difference between sound and language, between a signal indicating that you take action and a more complex message that communicates the situation, what you need to do, and why you need to do it. Systems designed around mere audibility are not enough to ensure intelligibility, which requires a more sophisticated approach to audio design. Voice systems will fail their intended mission if they cannot be readily understood by the target audience. The system will fail to get people to do certain things, such as shelter in place, if the message is poorly worded, too long, or fails to give specific commands.
An informal survey of leading ECS experts helped identify 10 common problems affecting voice quality and the effective use of ECS, issues that are not directly addressed by requirements in NFPA 72. While the body of NFPA 72 does not contain specific information on the design or evaluation of voice systems, the code’s Annex D does provide tips for good voice system design practices as well as a detailed system-testing protocol. A read-only version of NFPA 72 can be accessed using the online document information page at www.nfpa.org/72. Additionally, the National Electrical Manufacturers Association publishes a guide, SB 50-2008, Emergency Communications Audio Intelligibility Applications Guide, available at nema.org, which addresses some of these design issues.
1. Audibility: Not the same as intelligibility
Many designers, installers, and authorities assume that if the voice message is audible, it will be understood. A voice system must certainly be audible to be understood; most people have experienced paging systems that seemed to whisper and could not be understood over the background ambient noises. However, you might also have experienced a system that was too loud. Voice messages that are too loud can be distorted by overloaded system electronics, and often result in excessive reverberation in the space.
Even when a voice message is audible and presented at a comfortable listening level, it is not necessarily intelligible. A system that is intelligible is one that is clear, comprehensible, and capable of being understood. Imagine the phrase "Don’t use Stair B." Standing in a room, you’d hear the sentence come directly from the nearest loudspeaker. Fractions of a second later, the phrase would also come from the next-nearest loudspeaker. And fractions of a second after that, the phrase has bounced off the wall, ceiling or floor and is reaching you out of synch with the other sources. This reduces intelligibility, usually through the loss or corruption of consonants in the words. In this example, the "n’t" in don’t might be lost, leading to the message being interpreted exactly opposite of how it was intended. Or, the letter "T" might sound like "E".
Listeners should not receive sound from more than one source unless they are timed to arrive at the ear at the same instant. Effective voice system design requires that reverberation be minimized, and this can be done, in part, by not overdriving the system—by not making it too loud.
2. The quantity and spacing of speakers
Many system designers design a space using the same number of speakers as they would horns for a basic tone-only fire alarm design. Or they simply use combination speaker-strobes wherever a strobe is required. Neither method addresses the real factors that affect speech intelligibility.
If your ear is close to a sound source, the source does not need a lot of energy to be audible. A good analogy is head phones, which deliver a small amount of sound energy directly to your ear. Even when the volume is turned up to where you perceive it as being loud, those near you might not hear it at all. This analogy works well for most in-building voice system designs: Use more speakers, spaced closer together and driven at lower wattage levels.
How many speakers are required? And at what spacing and power level? It depends. A good design might start with the goal of having a uniform sound level where the listener never experiences more than about a 6 dB variation as they move about a space. This is a target used by engineers designing sound reinforcement for meeting rooms and some paging systems. An emergency system can usually tolerate a greater variation, provided it overcomes the background noise and provided it is not so loud as to create reverberation off the surfaces.
The sound pressure level must be enough to overcome most background noises, but not to the point where it’s judged "loud." For most occupancies, the level can be based on the ambient noise level measured at about 2000 Hz, a frequency that is an important component for speech intelligibility, particularly for consonants. A loudspeaker’s output varies with frequency and also varies as you move off axis—both will affect the required spacing. Additionally, a higher ceiling might actually require fewer speakers than a lower ceiling. However, because the speakers on a high ceiling are farther from the ear, they might require a slightly higher dB output, adjusted by using a higher power tap on the speaker or by using speakers with a different rating. Annexes A and D of NFPA 72 have diagrams and some discussion of these principles.
The 2010 edition of NFPA 72 includes a new tool/requirement for designers to designate Acoustically Distinguishable Spaces (ADSs). These are spaces that differ from others because of their acoustics, physical configuration, occupancy or system design. Establishing ADSs focuses designers and authorities on the possible need for different design principles to be employed.
3. The placement of speakers
Wall or ceiling: that’s the question. Ceiling-mounted speakers might be easier and less expensive to install and move in situations that have open, exposed installations or that have suspended ceilings. Wall mounting might help place speakers closer to the listener and reduce power requirements.
It’s easy to design for coverage at ear level when you know the characteristics of a speaker. Every speaker produces a cone-shaped output of audible sound, and the size of that cone can vary due to a variety of factors, including those described earlier. In reality, every speaker has output at some level beyond the cone described by the speaker’s characteristics. The problem is that the level can be quite a bit lower at some angles and varies with frequency—both factors that affect the quality of the speech intelligibility. Nevertheless, it might be acceptable to design certain spaces with reduced intelligibility, particularly in corridors where the occupants are generally mobile and can move short distances to areas of higher intelligibility.
4. The quality of a prerecorded message
Prerecorded message quality can be controlled better than live microphone announcements. Prerecorded messages should be carefully scripted and recorded by professional announcers/talkers who know how to use voice inflections, pauses, and enunciation to convey meaning.
The quality of a recorded message is also greatly affected by the control unit’s memory chip size and the specifications (bit depth and sampling frequency) used for recording. Emergency communications systems do not need the high fidelity and large file size used for, say, recorded music, but in areas with high noise levels or challenging acoustics, a higher-quality recording might be the difference between an intelligible system and one that requires considerable time and effort for a listener to understand—assuming the message is repeated enough.
In these situations, quality can be improved by using 16- or 24-bit depth versus the usual 8-bit depth, and by using a sampling frequency of at least 8,000 or 16,000 Hz. Sampling frequency has a direct effect on the consonants that are so important for understanding words. The sampling frequency needs to be at least two times the highest frequency that you want to reliably reproduce. So a sampling rate of only 4,000 Hz might save some chip memory, but would limit reproduction to sounds no higher than about 2,000 Hz. With that frequency limit, the voice message would sound flat and the consonants would be muddy.
5. Wiring and power
It is not uncommon to see 18- or 16-gauge wire for speaker circuits that are hundreds of feet long—the designers who created those circuits probably used a larger 10- or 12-gauge wire on their home theater speakers. Because audio circuits are alternating current, it is common to measure power loss in decibels rather than in voltage percentage, as is done for direct current fire alarm circuits. Calculations should be done by the installer or manufacturer to select a wire size that will limit the power loss to no more than 3 dB.
No design is perfect. Most will require adding a few speakers or changing power taps to adjust loudness, up or down. Wire sizes should be used that will allow additional load, and additional amplifier power capacity should be included to allow for changes and adjustments that might be needed to balance the system. Keep in mind that amplifiers can introduce distortion and noise when driven at their limits. This is another reason to increase the amplifier size beyond what the base design requires.
6. Emergency command center location and design
Architects and owners strive to make the best use of every square foot of a building. Providing a secure, fire-resistive emergency command center is usually not a priority unless it is required by a code or regulation. As a result, many emergency communications systems have their central interface, including a microphone, located in a building’s main lobby—usually one of the noisiest, least-secure areas of a building, particularly during an emergency.
The physical attributes of an emergency command center will vary based on the intended mission. Nevertheless, all command centers need to have low ambient noise levels to allow emergency teams to work and communicate. This is done by providing specific work areas for different functions and sufficient space for team members; think of the bridge on the starship Enterprise, from "Star Trek," with its specific duty stations, including one for the commander and one for a communications officer. Acoustic treatments need to be provided to absorb and dissipate conversational noise, and walls and service penetrations must be constructed to limit noise from the outside.
It is important that any microphone locations be positioned so that the user is not near others who must continue to talk. Also, there should not be a loudspeaker anywhere near the microphone location, which would cause feedback and noise in the system. The microphone cord should be long enough to allow a user to sit at or reach a desk or workstation where they might have drawings, operational plans, message templates, scripts, or other notes that they need to consult while making announcements. Placing sound-absorbing material above microphone locations has been shown to reduce noise and increase overall voice intelligibility.
7. System complexity andergonomics
People have come to expect intuitive, ergonomic user interfaces for computers, phones, music players, and household appliances. Similarly, the user interface for an ECS needs to consider the mission and the users. Systems that are used daily for routine functions allow users to become familiar with the controls and comfortable with the complexity of the system; permitting a voice ECS to be used for non-emergency purposes was a major step towards improved usability in the 2010 edition of NFPA 72. Systems that are only rarely used, on the other hand, require simpler interfaces. Emergency forces like police and fire might have personnel who have been trained and are capable of using the system interface. In other situations, it might be necessary for owners to have qualified persons available to assist or to issue announcements at the direction of the emergency commanders.
The characteristics of the system microphone are important ergonomic factors that affect voice intelligibility. Some microphones need to be held close to the mouth, perhaps an inch or less. Others need to be three or four inches away. How is the user to know what’s ideal? A simple diagram next to the microphone can help. Some microphones are very directional and must be held flat in front of the speaker’s mouth. These microphones are useful in small command centers, since they’re less likely to pick up conversations off to the sides. On the other hand, microphones with a wider polar sensitivity are more forgiving for a user to hold comfortably while moving and doing other tasks. Their downside is that they will pick up extraneous noise in poorly designed command centers.
8. When and how to test voice systems
Annex D in NFPA 72, prepared with the help of the Fire Protection Research Foundation, describes detailed test protocols, including information on how to plan tests. The test protocols in the annex are not required; the code permits them to be used, but also allows a simple "listen" test.
NFPA 72 requires loudspeakers to be tested at the time of acceptance and once per year. The testing, however, is very different from tone-only systems, since voice system intelligibility is affected by more than just audibility. Measuring the audibility of a voice message using a sound meter is virtually meaningless with respect to intelligibility, since the presence of furnishings, carpet, and people can drastically alter the quality of voice signaling by absorbing different sound frequencies. In many cases, though, furnishings and people can improve voice intelligibility by reducing reverberation. Also, because different noise frequencies affect different voice sounds, called phonemes, it is important that the expected noise be included as part of a test.
Intelligibility meters can also be used to measure system performance. A special sound that contains all the phonemes that make up human language is played through the system, with the meter scoring the delivery. The test sound can be prerecorded on the voice chip by the system manufacturer. The test protocol in NFPA 72 also includes a method to include the microphone in the testing. Using the microphone is an opportunity for people to test how to get the best voice quality. It is also an important test of one more piece of electronics that can dramatically affect voice quality.
An additional consideration with prerecorded messages is that the quality of those messages may not be as critical as the quality of live announcements, because prerecorded messages are usually automatically repeated several times, giving listeners the chance to resolve questionable word sounds. Research has shown that if you understand about 80 percent of the words, your sentence comprehension will be in the high nineties because your brain is so adept at putting things in context. By repeating a message several times, the accurate receipt of the message is almost guaranteed, except in the worst conditions. But a message spoken by a commander into a microphone might not be repeated at all, or it might be repeated using different words or sentence structure.
Without the verbatim repetition, the acoustic environment and all of the pieces of equipment in the chain, including the microphone, become more important to speech intelligibility and accurate message receipt.
9. What the voice message should say
You can have the best sound system ever made, but if you don’t say the right things, you won’t get people to do what you want them to do. Worse, you might do harm.
A message will not be understood if the person talking has an unfamiliar accent, talks too fast, holds the microphone too close or too far away, or uses slang or complex language. One expert pointed out that the word "please" should not be used in prerecorded messages; emergency announcements need to be clear, direct, and stripped of all unnecessary language. Messages have two main purposes: To inform people of what’s going on, and to direct their behavior.
They should contain three or four critical elements: what has happened, what you should do, why you should do it, and "who am I"—what authority is telling you this. Note that "what you should do" should be the last element spoken, since it will be the best remembered. Example: "There is a fire on floor number 15. For your safety, the fire chief wants you to evacuate using the stairs."
There are many other factors that affect good messaging strategies. The Fire Protection Research Foundation is working with the National Institute of Standards and Technology on behalf of the NFPA 72 ECS Technical Committee to develop guidelines and templates for a variety of emergencies, target audiences, and delivery platforms, including voice communications.
10. Who will be allowed to use the system?
Answering the question of who will authorize and make announcements requires careful planning and discussion among all of the stakeholders involved in ECS planning and implementation.
Systems with prerecorded messages can be triggered automatically for emergencies such as a fire, where scenarios have been developed and analyzed, and where the needed actions are well established. Even where prerecorded messages have been automatically triggered and broadcast, however, the issuance of live messages by the emergency team will improve effectiveness; in some situations, prerecorded messages can be judged irrelevant by occupants, just as tone-only fire alarm signals are often ignored. Also, many emergencies will almost always require some assessment, decision making, and customizing of message templates before issuing voice announcements.
In a fire scenario, live announcements might wait until incident command has been established, and until it has had an opportunity to gather critical information that might affect the message content. However, when there is a person with a gun in a classroom, more immediate use of the ECS might be necessary. That’s why the question of authorization needs to be a part of the emergency plan. The issue of access, both physical and/or password control, to the ECS controls and microphone must be worked out before a system is designed and installed. Similarly, where there are multiple microphones or command stations, protocols for control, access, and priority need to be established.
An ECS requires careful planning, design, installation, and use. Systems that rely on voice for message delivery face numerous challenges that involve many different individuals, authorities, trades, and professions. It is important that the relevant stakeholders and experts be identified and involved early in the planning of any ECS project. Because voice system design is so different from conventional fire alarm signaling design, engineers need to learn new techniques and use new design tools or seek partnerships with experienced professionals. Authorities and owners need to be actively involved in the planning of these systems; ignoring ECS voice issues or only partially addressing them can jeopardize the quality and effectiveness of voice communications during an emergency.
Robert Schifiliti, a licensed fire protection engineer, is president and CEO of R.P. Schifiliti Associates, Inc. He is involved with a number of NFPA committees, and chairs the NFPA Signaling Systems for the Protection of Life and Property Technical Correlating Committee responsible for NFPA 72.