Examples of speaker verification systems

   

Technology providers usually offer speaker verification  sytems that could be integrated in products. In this section, we examine a few examples.

• VOCALIS Ltd. (at that time part of LOGICA) developed a real-time   telephone-based text-dependent system  in 1988. The system's decision making logic has four layers of authentication prior to making the final decision on the claimed identity of the speaker. Individual password and PIN are used as the first two layers of authentication. A number of words from pre-defined vocabularies are also chosen randomly at the later stages of verification. The enrolment process follows a robust strategy, in order to guarantee that the reference template is relevant. VOCALIS' speaker verification system is developed for a UNIX platform  and makes use of embedded hardware. It is fully integrated into the company's communications infrastructure. An extensive field  trial of the system was organised in conjunction with a telephone service company in the US, where over 400 people from four different geographical groups registered with the system. The system allowed users to access US Government databases out of office hours. All registered participants were encouraged to make at least ten calls over a three month period, using as many handsets as possible. The registered speakers  were also asked to make two impostor  calls and were purposely provided with information to break the first two layers of authentication. At the end of the trial, more than 4000 calls were logged. A detailed analysis of the results showed a 5% equal error rate  purely for the verification algorithm. None of the participants managed to break into the system as impostors  when all layers of authentication were active. A similar speaker verification  system is available also on the Callserver platform from VOCALIS.
• ENSIGMA Ltd. proposes a speaker verification  system named VERIFIER [Moody (1991)], which runs on the Loughborough AT&T DSP32C telephone board. It works both over telephone line and audio line. ENSIGMA claims full portability  of the speaker verification  software on other boards. The system uses a new approach to the speaker verification  problem. In fact, instead of simply matching a speaker's voice against the user's template, ENSIGMA's system compares the spoken word also against a ``general world model''. The decision is made taking into account the results of these two comparisons. This method has been evaluated over the telephone line and gives 1% false rejections   and 1% false acceptances . The method also seems quite robust to background noise . Enrolment consists of the new user repeating a series of digits; the duration of this phase is about one minute. The access consists of the pronunciation by the user of a list of (random) digits requested by the system. System response takes about 5 seconds. The VERIFIER has many properties that make this system a good candidate as a possible reference hardware system for (text-dependent) speaker verification . The multilingual aspect, or generally speaking the vocabulary  problem, may be solved only if procedures to build the ``reference world model'' used to score utterance similarity is released by the producer.
• The CNAPS of Adaptive Solution Inc. works over the telephone line and gives both speaker verification  and speaker identification  capabilities [Skinner et al. (1992)]; averaging the true acceptance  and true rejection  rates the system reaches 95% of correct responses for speaker identification  and 97% for speaker verification . These results refer to a field test  where the user pronounces the seven digits of their PIN to access a system; for speaker verification, the user communicates a reference number using DTMF  to the receiver before starting to pronounce the seven digits.

  Voice Control System offers speaker verification  in their speech recognition system  over the telephone line. It works on several platforms including PC   and VME . By August 1990 [Hunt (1991)] the performance was 1% false acceptances  and 2% false rejections.  

  A speaker verification  facility is also part of the AT&T HuMaNet teleconferencing system [Berkley & Flanagan (1990)] that works via ISDN  and uses an AT&T digital signal  processor DSP32C board on a Personal Computer .

The main problem which service providers and integrators have to face is to be able to predict what level of performance and user satisfaction they are going to get from a given technological solution.