Voice-ID Technology Promises Secure Systems
(08/11/98; 2:00 p.m. ET)
By Stephan Ohr, EE Times
Keyware Technologies, Brussels, Belgium, and STMicroelectronics have jointly announced plans to collaborate on the development of a voiceprint identification system that implements Keyware's Layered Biometric Verification (LBV) technology and a specialized digital signal processor (DSP) chip from STM.
The system will use a spoken code word rather than a personal identification number to identify a unique user, which could make smart cards, admission keys for secure facilities, and credit/debit cards more secure. "If a robber steals your badge, you still must be there to open the door," said Paolo Gonella, the director of the automotive multimedia business unit of STMicroelectronics.
Voice-based security is just one application that uses the TDA7551 DSP core developed in Agrate Brianza, Italy by STMicroelectronics. Others include admission to secure parking garages and automotive command-and-control applications, said Gonella. In fact, STM has licensed voice recognition software from Lernout & Hauspie, Burlington, Mass., for car information systems, said Gonella.
Biometric voiceprint identification, however, promises to be a dramatic applications of the DSP device. Keyware refers to this field as "biometric verification and authentication," a relatively new discipline that uses electronic sensor data to confirm the unique identity of a person. Biometric verification systems are essentially data-acquisition systems that read analog sensor inputs, convert them to digital data, and use a processor to manipulate the data and compare it with reference data in a lookup table. In the case of speech verification or voiceprint identification, amplitude, frequency, and other characteristics must be extracted from voice input. This extraction and comparison is performed by a DSP.
The voiceprint ID system developed by STM for Keyware uses the TDA7551 50-million-instructions-per-second DSP core and a specialized speech-coding algorithm that speeds identification. The chip is responsible for both voice "training" and verification, said Umberto Zinghieri, a speech-application engineer in STM's automotive multimedia business unit. In training, the user must repeat his or her password three times for the voice features to be extracted, encoded, and recorded on a smart card. The verification technology features ultra-precise pattern recognition.
The TDA7551 includes an internal analog/digital and digital/analog converter. A 16-bit device is satisfactory for voice recording, said Zinghieri, though 20 bits is now commonly used for hi-fidelity audio. The voiceprint codec and DSP will be resident in the electronics of a bank automatic teller machine or automobile. In operation, the system will listen to a spoken password and compare its extracted features with the data previously recorded and stored on either a user's smart card or in a large multiuser database.
Since the TDA7551 includes a digital interface to the smart-card reader, a car- or ATM-based voice recognition system can be implemented with few external components, according to STM. Requirements include a microphone preamp, perhaps a set of light emitting diodes to indicate verification status, and for large multiuser installations, a large memory space. A single voiceprint password composed of a two-second sound would require about 11 kilobytes for each user, said Zinghieri. This can be stored on a user's smart card. For multiuser systems, 4 megabytes of memory would store voiceprints for about 400 users.
For bank ATMs, parking garages, and automobile consoles, the embedded systems requirement would force STM to market the device for a very low price. Though STM refused to comment on pricing, Zinghieri did not refute speculation that the device would be priced below $9 or $10. The company did say the parts are custom circuits whose specific performance features are subject to customer preferences. STM intends to offer evaluation samples of the circuits to customers in the fourth quarter, and to start volume production in the second quarter of 1999.
Keyware has had many trials with sometimes inconsistent results, according to Zinghieri. There is also a concern that a tape recording of a user's voice may trick machinery into mistaking the recording for the actual user. But Zinghieri believes the combination of STM's processing power, along with specialized algorithms and Keyware's software, would substantially reduce potential errors in identification.
Biometric Identification In The Future
But voiceprint identification is only one aspect of biometric identification. Future systems will include visual images such as a user's face and fingerprints, said Zinghieri. In fact, STM demonstrated a capacitive fingerprint reader at a press briefing in May.
"Layered biometrics can be used to restrict physical access, to verify identity for financial transactions, and to control access to information over computer and telephony networks," Keyware said. Keyware's LBV Security Server, in fact, is actually a specialized user database, which in principle, records and synthesizes many types of data extracted from users. In the future, spoken passwords -- combined with visual-pattern recognition and touch-sensor data -- can reduce the possibility of misidentification down to zero, the company said. <Picture: TW> |
