Phonetic technology provider Nexidia will announce on Tuesday Language Assessor, an application that can test the verbal proficiency of non-native speakers.
Nexidia uses what it calls Phonetic Search Engine (PSE) technology to do phoneme pattern matching that can find individual words in a broadcast.
A phoneme is usually defined as the smallest phonetic unit in a language that is capable of conveying a distinct meaning, such as the m in mitt and the b in bit.
Language Assessor will use the same PSE technology to test speakers by identifying and rating the clarity of their pronunciation of words, phrases, proper names, slang, and nonstandard grammar patterns.
Because there are a finite number of phonemes in any speech, Language Assessor is easily ported to languages other than English.
The target market for Language Assessor, also known as an accent analyzer, are companies considering offshore tech support sites where English is either not native to the population or the spoken English accent might be difficult for a North American native speaker to understand.
Elizabeth Herrell, both a contact center and voice communications specialist for Forrester Research, said this will be a useful tool and technology for companies who use agents where English is their second language.
Up until now, many companies make the mistake of giving only reading and writing tests, not realizing an agent can achieve a high score in written skills while at the same time speaking heavily accented English.
In addition, Herrell noted that some companies may want to use this assessor to offer levels of customer service. Dubbed "right shoring" or "best shoring," these kinds of service-level offerings typically connect premier customers needing help to native speakers, while individuals may only get to talk to agents located offshore.
The Language Assessor gives offshoring companies pitching a company for a customer service contract a script that is read by candidates and recorded.
The program takes the recording and searches for words and phrases used in a predefined script.
The second part of the assessment matches the speaker's phonetic pronunciation against the North American English phonetic model.
The third component measures the pace at which the script is read.
Following the tests, an aggregate score is given, allowing a company to select only the top performers, said Anna Convery, a senior vice president and product manager at Nexidia.
The problem with using live language screeners, according to Convery, is that they may be strict at first, but as time goes on, they tend to relax and get listener fatigue.
Currently the company offers 33 language models.
The audio search engine can index one hour of audio in 17 seconds, after which an actual audio search would take a fraction of a second.
According to Herrell at Forrester Research, there is currently no other company offering a similar language assessment capability.