Measuring and improving speech intelligibility with speakers

The purpose of a sound system is the transmission of acoustic information (speech and sound). In the case of announcement systems, voice alarm systems and public address systems, the goal is to transmit comprehensible speech to the audience and receivers of the message with as little loss as possible. This aspect is then far more important than the sound quality itself, because it makes no sense to develop a system that nobody can understand or that doesn't get the message across. Although sound quality and speech intelligibility are inseparably linked, they are not the same thing. It is possible that

• a poor-sounding system with high speech intelligibility (e.g. a frequency-limited horn speaker with an uneven frequency response) will only deliver a very mediocre musical sound;
• a top-quality low-impedance speaker will be practically incomprehensible for voice announcements. For example, if you were to operate it as a single speaker in a large factory building. At the same time, this low-impedance speaker can deliver music reproduction of excellent quality.

Although there have been many attempts to objectively measure speech intelligibility, the most commonly used parameter is the Speech Transmission Index (STI) and its derivatives. The STI is based on the relation between perceived speech intelligibility and intensity modulation in the speaker's voice. The STI method is described in the IEC 60268-16 standard .

In short, the STI measures how close the transmitted sound remains to the original sound wave

STI values range from 0 to 1, the higher the better. If the sound transmitted to the ear contains many more reflections than the original, the STI will be worse. This is because the transmitted sound then differs greatly from the original. Conversely, the more similar the sound arriving at the ear is to the original sound, meaning that it contains less reflected noise, the higher the STI. As with all objective electroacoustic measurement methods, the STI does not measure the actual intelligibility of speech, but only certain parameters that strongly correlate with intelligibility.

The key factors of speech intelligibility are:

• Bandwidth and frequency response of the PA system

• Volume and signal-to-noise ratio (S/N)

• Reverberation time in the space

• Spatial volume

• Shape of the space

• Distance between listener and speaker

• Polar pattern of the speaker

• Number of speakers running

• Ratio of direct sound to reverberation (directly dependent on the last five factors)

• Speaking rate of the speaker

• Hearing ability of the receiver

Secondary factors of speech intelligibility include:

• Sex of the speaker (different voice frequencies)

• System distortion

• The equaliser settings of the PA system

• Uniformity of coverage

• Sound focusing and presence of discrete reflections

• Direction of the sound arriving at the listener

• Direction, frequency and volume of disruptive noise

• Vocabulary and context of the language

• Microphone technology of the speaker

• Manner of speaking of the speaker

Some of these parameters are building-related or system-related. Other parameters depend on human factors and are beyond the control of the system. Please also note that some important factors (how the speaker speaks and how the receiver listens) are beyond the control of the system and the architect. Psychoacoustic effects also contribute to the loss of speech intelligibility, especially at very high vocal levels. Low-frequency components in speech can mask quieter, higher-frequency sounds and thus limit their perception. Yet another kind of interference are linear and non-linear distortions in the signal chain. These distortions come not only from overwhelmed PA amplifiers and bad speakers, but also from well-intentioned yet poorly implemented signal processing. Excessive compression, overwhelmed limiters, or unnecessary increases or decreases in the frequency range due to equalisation may impair speech intelligibility.

The following tips improve the speech intelligibility of a PA system:

• Point the speakers toward the audience, keeping as much sound away from the walls and ceiling as possible.

• Ensure a direct line of sight between the speakers and the audience.

• Ensure that the speakers have sufficient bandwidth: at least 250 Hz to 6 kHz, preferably up to 12 kHz

• Avoid frequency response anomalies. Roll out the bass and ensure an adequate, but not excessive, high frequency range.

• Avoid mounting speakers in corners.

• Minimise the distance between the speaker and the audience.

• Ensure a speech S/N ratio of at least 6 dBA, preferably > 10 dBA.

• Make sure that the microphone user has been properly trained and understands the need to speak clearly and slowly in reverberant environments.

• Place the announcement microphone in a quiet area or a shelter. Use an effective close-talking microphone with squelch.

• Use a microphone with good (steady) frequency response.

• Avoid long delay times (> 50 ms). Use electronic delays and speaker spacing of less than 15 metres.

• Use automatic noise level measurement and gain adjustment to optimise the S/N ratio.

• Use highly directional speakers in reverberant spaces to optimise the D/R ratio. Use models that have flat or consistently controlled acoustic power consumption.

• Minimise variations in direct radiation behaviour. Fluctuations as small as 3 dB can be detrimental in dynamic spaces with a lot of movement.

• Consider improvements to the acoustic surroundings. Plan the PA system so that it is not isolated; remember that the acoustic surroundings limit the performance of any PA system.

• In highly difficult conditions, use simple vocabulary and short compact message formats.