“Atos Origin optimizes Speech Recognition with its noise filter system”
Paris, 11 June 2002
Atos Origin is introducing new VSAD* (Voiced Signal Activity Detector) software that enhances the quality of telephone services using speech recognition by suppressing un-voiced signals.
Most current telephony-based speech recognition solutions are activated by the detected signal, without distinguishing between useful signals (speech) and background noise. As a result, users have to pay higher operating costs and bear the inconvenience of disturbances.
To overcome these difficulties, the Research & Development Department of Atos Origin's Multimedia Division has developed VSAD software to detect unvoiced signals and filter a significant portion of the background noise, such as breathing, sneezing, coughing, and crackling. The system enhances the speech recognition performance of telephone services, especially when they are consulted in high mobility conditions or in extremely noisy environments, such as public areas, transportation or construction sites.
> VSAD selects voice signals and suppresses background noise.
Only useful signals (containing speech) are input to the speech recognition system, while any non-speech signals are identified and directly eliminated. The VSAD technology detects the signal frequencies (pitch*) generated by human vocal cord vibrations.
> Intelligent barge-in*
The barge-in function allows the user to interrupt the voice telephone service. Without VSAD, background noise from various sources could interfere with the communication by causing untimely interruption of voice server message output. With VSAD, the barge-in only processes the voiced signals, thus ensuring unparalleled user comfort.
> VSAD adapts the speech recognition system to background noise.
Unlike the traditional VAD*, which is quickly defeated by an overloaded sound signal, VSAD enhances the quality of service by adjusting to low signal-to-noise ratios of less than one-tenth of a decibel. This level is compatible with the use of a voice service in a car with the windows open and driving on an highway, or with a person mumbling into the telephone.
Stand-alone VSAD Software
Developed by Atos Origin, VSAD is independent of speech recognition engines. When installed on the front ends of voice servers, its resource requirements are virtually negligible, thanks in particular to use of the MMX functions of Pentium© processors.
Functional and Cost Benefits
- Background noise filtering minimizes the volume of input data to the speech recognition system. Enhanced quality of service is guaranteed by the reduction of communication interruptions caused by background noise. In addition, certain malfunctions due to misinterpretation of background noise are eliminated.
- Production costs are optimized as a result of the lower speech recognition resource requirements, since resources are only used when necessary.
VSAD: Voiced Signal Activity Detector.
VAD: Voice Activity Detector (a simple sound energy detector).
Pitch: Vibration frequency (of human vocal cords in this case).
Barge-in: Ability to interrupt voice server message output.
MMX: MultiMedia eXtension (Intel©).
About Atos Origin
Atos Origin is a leading international business and technology integrator. Its business is turning client visions into results through the application of management consulting, enterprise, e-business and outsourcing solutions. The company has annual revenues of EUR 3 billion, operates in more than 30 countries worldwide and has over 26,000 employees. Atos Origin's clients include ABN-Amro, Alstom, BNP Paribas, Euronext, Fiat, ICI, KPN , Lucent, Philips, Renault, Saudi Aramco, Shell, Unilever and Vivendi Universal and Wolters Kluwer.
Voice Technology Expertise
Atos Origin's Multimedia Division has been expanding its expertise in interactive voice services since 1988 on an industrial-scale multimedia platform open to third-party software, such as speech recognition or voice synthesis engines. In addition to VSAD*, Atos Origin's voice technology expertise includes voice synthesis, VoiceXML, authentication by voice signature, multimedia contacts centers (using virtual ACD and CTI), and, thanks to its Research & Development investment, the full range of emerging voice technologies.
Atos Origin Press Contact:
Anne de Beaumont
Tel: + 33 (0)1 49 00 96 42