Speaker Verification and Identification

Speaker Recognition can be classified by function in:

• Speaker Verification – the process of accepting or rejecting the identity claim of a speaker;

• Speaker Identification - the process of determining which registered speaker provides a given utterance.

In speaker verification, a voice print of an unknown speaker who claims an identity is compared with a model for the speaker whose identity is being claimed. If the match is good enough, the identity claim is accepted. A high threshold reduces the probability of impostors being accepted by the system, increasing the risk of falsely rejecting valid users. On the other hand a low threshold enables valid users to be accepted consistently, but with the risk of accepting impostors. In order to set the threshold at the optimal level of impostor acceptance (false acceptance) and customer rejection (false rejection), data showing impostor scores and distributions of customer are needed.

In the speaker identification task, a voice print of an unknown speaker is analyzed and then compared with speech samples of known speakers. The unknown speaker is identified as the speaker whose model best matches the input model.

One of the differences between identification and verification is the number of decision alternatives. In verification there are only two choices, acceptance or rejection. In identification, the number of decision alternatives is equal to the number of population. Thus, while speaker verification performance doesn’t depend on the size of the population, speaker identification performance decreases as the size of the population increases.

It should be mentioned that there is also a case called “open set” identification, in which a reference sample for an unknown speaker may not exist.

The efficacy of speaker verification systems can be measured by using the receiver operating characteristics (ROC) curve adopted from psychophysics. The ROC curve is obtained considering two probabilities:

• the probability of correct acceptance (false rejection rate);

• the probability of incorrect acceptance (false acceptance rate).

These probabilities are assigned to the vertical and horizontal axes respectively, varying the decision threshold.

Another measure of system performance is the equal-error rate (EER). It corresponds to the threshold at which the false rejection rate is equal to the false acceptance rate.

Speaker verification is generally used in applications which require secure access. The systems operate with the user's knowledge and usually require their cooperation.

Speaker identification systems generally operate without the user's knowledge. It can be applied in routing users to the correct mailbox, identifying talkers in a discussion, alerting speech recognition systems of speaker changes, checking if a user is already enrolled in a system, etc. Speaker identification is also used for police purposes. For instance, a criminal's voice can be cross checked against a database of criminals' voices looking for a match, and ergo the identity.

Speaker identification problems consist in:

  • Verification Distinguishing multiple speakers during a conversation;
  • Identifying an individual's voice based upon previously supplied data regarding that individual's voice.

Speaker identification is based on complex voice processing algorithms. In contrast, speaker verification is based on more simple voice print comparing.

 

Latest News

Stan Billue releases his classic Sales Training Series in MP3 Format

June 09, 2009 by Stan Billue

Although recognized as one of the top Telemarketing Sales Trainers in the World, Stan Billue is the first to admit that he is rather slow on adapting to new technologies. Stan stated; “I was probably one of the last professional Speakers to add a Fax Machine, a Computer and then a Cell Phone. I also drug my feet at converting my Audio Series from Cassettes to CDs a few years ago.”

Read more...
 


Contactual Partners with Pipkins to Provide Workforce Management to Contact Center Customers 

June 04, 2009

Joint Marketing Partnership Brings Together Best in Class Workforce Management and Hosted Contact Center Software. 

SAN CARLOS, CA, June 04, 2009 /24-7PressRelease/ -- Contactual (http://www.contactual.com/pr), the global leader in on-demand contact center software that empowers customer facing organizations to end customer frustration, and Pipkins, Inc. (http://www.pipkins.com/) a leading supplier of workforce management software and services to the call center industry, today announced they have joined in a partnership to cross sell technology. This agreement will provide a direct connection for call center managers looking to combine a trusted workforce management system with hosted contact center software.

Read more...
Direct-Inbox