Recent articles

The problem

Many of us work in offices, often with meeting rooms adjacent to each other or directly adjacent to open-plan areas. This requires speech privacy to be considered, for example, how private a conversation in one space is from people in the neighbouring space.

There are situations where the conversation may contain very sensitive information. In these situations there is a need to quantify exactly how much information is leaked via casual overhearing or, in specialist cases, how it may be deliberately intercepted by a covert listener. It is this quantification of information leakage which allows the speech security – and in turn the risk involved – to be assessed.

Speech privacy generally considers disturbance from the audibility of speech. However, speech security is considered different from speech privacy as is involves the assessment of risk of information communicated via talking being leaked.

Previous work

Much work has been carried out into speech privacy. Work by Gover and Bradley [1] looks at using new descriptors to better relate speech security to thresholds of intelligibility, cadence and audibility. Research by Bradley and Gover [2] was used as the basis for the American Standard ASTM E2638-10 [3] and combines sound insulation ratings with background noise levels to give a speech privacy class, though it does not explicitly set values for adequate speech privacy. Additionally, the standard sets out how many words would be expected to be intelligible over a given period, yet it is not clear how many words have been spoken so it is not possible to derive a percentage of words identified.

New research

The new piece of work [4] aims to give a meeting room a single rating which conveys the expected speech security and takes into account:

• vocal effort level of the talker
• reverberation in the meeting room
• sound insulation provided by the separating elements (ie, door, wall, etc)
• background noise at the listening position.

The single rating most relevant to those using a meeting room is how loud they can speak for a specific level of speech security; this is maximum safe vocal effort level (MSVEL). Given people can talk with a range of vocal effort levels (eg, normal, raised, loud or shouted), MSVEL tells a talker how loudly they can speak for a given level of speech security.

The MSVEL has to be related to measurable parameter that takes into account the issues listed above. Conveniently, the speech transmission index (STI) parameter does this, though the major challenge to be overcome by the research is relating MSVEL to STI. The aim is to calculate an offset, in decibels, that relates a measured STI of the transmission path (ie, 0.1, 0.2 or 0.3) to a MSVEL for percentage of words identified by the listener (ie, 1%, 2% or 5%). This is achieved by carrying out a series of decision-based experiments where participants listen to sentences with varying vocal effort levels and the meeting room the listening position transmission path simulated using signal processing techniques [5], and have to identify the words being spoken. The experiments quantify offsets which link the STI of a transmission path to a specific percentage of words identified by the listener outside the meeting room.

Engineer holding a sound level monitor

The decision-based experiments were carried out with male and female talkers and showed that the offset for male talkers is approximately 10 dB larger than for female talkers, hence setting an MSVEL based on male talkers could be considered a ‘worst case scenario’. Based on this, offsets are provided which link a percentage of words identified to an STI, these can then be used with the in situ method presented for determining MSVEL for a meeting room.

This work effectively shows how an MSVEL can be set for existing meeting rooms which allows the loudness of the talkers to set based on the level of speech security needed.

View our office experience or contact us to find out how we can help with your project.