NEC Technologies Enable Effective Voice Control of Electronics in Noisy Environments

*** For immediate use May 7, 2013



Tokyo, May 7, 2013 - NEC Corporation (NEC; TSE: 6701) has developed speech recognition technologies that enable household electronics to be controlled by voice commands through a combination of the company's noise cancellation and speech enhancement technologies.

In recent years, technologies have enabled information devices, such as electrical appliances, tablets, and smartphones, to be controlled by voice. However, these devices require built-in microphones that must be spoken to at close range. If users in a noisy environment are slightly removed from a speech recognition capable device, whether watching television, preparing a meal, or riding in a vehicle, the device tends to respond to background noise, rather than a user's command, which often results in mistaken operations.

NEC's latest technologies enable devices to be reliably operated by voice through an interface and microphones that accurately recognize a person's voice. This accuracy is maintained even when users are distant from a device in a noisy environment, where existing technologies have difficulty operating effectively.  As a result, users who are in noisy locations, such as a busy living room or kitchen, can still operate electrical appliances and specialized devices by voice when separated by roughly one meter (approximately 3 feet).

"These innovative technologies can enable car navigation systems to overcome background noise and accurately detect voice commands from the driver of a moving car," said Akio Yamada, General Manager, Information and Media Processing Laboratories, NEC. "Employees of warehouses, manufacturing plants and retail stores can also benefit from the convenient data input of these new technologies as their hands remain free to focus on other tasks."

Primary features of these technologies include:

1. Accurate voice detection using noise cancelation technologies with two microphones
   These newly developed technologies enable devices equipped with two microphones to detect voice and background noise signals, then to sequentially estimate and remove the noise components, leaving only the voice component behind. Moreover, the technologies accurately follow and remove background noise as it changes over time, thereby detecting voice signals more accurately than existing noise cancelation technologies. As a result, electrical appliances and devices can be easily operated by voice while users' hands remain free.



2. Effective speech enhancement using models of speech from a quiet environment
   Following an initial noise cancellation, speech enhancement is carried out using models of speech from a quiet environment, which clarifies speech signals that continue to carry distortion and left over noise. As a result, the adjusted speech signals become suitable for speech recognition, and the accuracy of these new technologies becomes greater than the current world standard (*) for speech recognition.

Note

(*) Based on standards established by the ETSI Advanced Front-End (AFE) for advanced speech processing using a server-client framework (distributed speech recognition; DSR) that sends recognized data back to a terminal once a server can provide recognition results for voice data received from a mobile terminal.