Research and Development

Speech Privacy

Speech Privacy

At work, at home, and in public, we often find ourselves in spaces where we all too easily overhear other people's conversations. This poses two major problems. First, there is the risk of leakage: for example, the risk that sound from an important and confidential conference may be overheard by someone in the hallway. Second, there is the risk of distraction: for example, where employees cannot avoid overhearing outside conversations that may cause them discomfort, impair their concentration, and reduce their productivity. These problems can be overcome by using techniques to promote speech privacy.

Sound Masking

Traditional sound masking = energy masking
Traditional sound masking works by playing out artificial air-conditioner noise at high volume, drowning out the sound of nearby conversations. It is therefore a type of energy masking. Although this method has found wide use, the noise that it produces can itself be distracting and a cause of discomfort.

Yamaha's alternative sound masking = information masking Our alternative information masking approach utilizes a masking sound created from human voices, and works by camouflaging conversational content rather than by drowning it out. The masker does not sound like noise, and is effective at lower volume than typical masking noise. As a result, our new approach causes much less discomfort than traditional masking.

The conceptual drawings above illustrate the difference between energy masking and information masking. Energy masking works by “painting over” ambient sound, covering it up. Information masking works by camouflaging surrounding speech sounds so that they cannot be understood.

Disruptive masking sound is generated from human voices
Yamaha's information masking technology uses our newly developed disruption masker to camouflage ambient(overhearing) speech. Created from a mix of recorded human voices, the masking sound mixes with and obscures the content of actual conversations in the environment. Ambient(Overhear) speech disappears into the masking sound in the same way that a chameleon blends into the background. Because the masking sounds like human voices, it does not cause the discomfort so characteristic of traditional masking techniques.

Toward a more comfortable environment…
Yamaha's information masking technology uses our newly developed disruption masker to camouflage ambient(overhearing) speech. Created from a mix of recorded human voices, the masking sound mixes with and obscures the content of actual conversations in the environment. Ambient(Overhear) speech disappears into the masking sound in the same way that a chameleon blends into the background. Because the masking sounds like human voices, it does not cause the discomfort so characteristic of traditional masking techniques.

Performance Evaluation

Information masking provided high-level protection to 90% of vulnerable area outside of the conference room, as compared to 0% protection when no measures were taken. People standing in this protected area will find it difficult or impossible to hear what the conference speakers are saying.
(In the illustration, the “highly protected” portion of the indicated area jumps from 0% to 90% when masking is applied.)

Developing a Practical System:
In a joint research effort, Yamaha and Institute of Construction Engineering are carrying out quantitative evaluation of the information masking technology, and reporting the results at academic conferences both inside Japan and abroad. These evaluations have conclusively shown that information masking sound effectively masks ambient conversations, and does so at lower volumes than traditional energy masking sound.

In the graph of experimental results at right, the vertical axis plots the speech intelligibility (the degree to which the targeted speech we are trying to mask can be understood), and the horizontal axis plots the loudness of that targeted speech relative to the loudness of the masking sound.. Higher values along the vertical axis correspond to greater intelligibility (less effective masking), while higher values along the horizontal axis correspond to louder targeted speech.

Assume, for example, that the required speech intelligibility is 0.4. The graph shows that, when the Yamaha masker and the energy masker are played at equivalent volumes, the Yamaha masker will obscure target speech that is 10dB louder than the energy mask can handle. This in turn means that Yamaha masking played at low volume will be as effective as conventional energy masking played at higher volume.

Return to Top