ЦРТ: Speech Analytics Lab

Speech Analytics Lab is tools of speech analytics.

A system is intended for the solution of search and analytical tasks on arrays of the unstructured voice information using sensing technologies of a continuous speech, the analysis of texts and the Data mining methods.

Speech Analytics Lab allows to carry out job analysis of contact center in absolutely new and most relevant section – by subjects of the client addresses defined by a speech sensing technology.

Screenshot of the software window, (2013)

This property defines the main importance of a system for work of contact center and for all business which tasks include active and sensitive work with audience of end users of products and services.

A system is focused on specialists of corporate services of quality (estimates of indicators of productivity of work of operators and quality by subjects of client addresses), analysts (identification of customer needs, search of points of growth of business, detection of non-core addresses), trainers (formation of thematic selections for preparation of training courses).

Functionality

• System operation of a thematic clustering does not require training of engines of speech recognition, selection of dictionaries of controlled lexicon or other technology procedures. A system will start the analysis of the voice information at once as soon as gets access to phonograms of calls.

• Sensing technologies of a continuous speech on the big dictionary (LVCSR – Large Vocabulary Continuous Speech Recognition) and information extractions using cluster analysis of data (Data Mining clustering) are the cornerstone of technology basis of a system. As a result of automatic recognition the speech of announcers will be transformed to the text indexed file suitable for the automatic lexico-semantic analysis.

• A system is simple in work, has the thought-over ergonomic interface which main objective to give a correct, complete and general idea of results of semantic analysis of huge arrays in thousands and tens of thousands of calls. Data view in the form of "a semantic cloud" or "tag cloud" is for this purpose used. The legend of "cloud" contains information on the power of a thematic cluster, frequency of use of the fundamental lexical units defining a subject.

• Management of "cloud" at the level of redefinition semantic a centrode of a subject, removal of separate words in the stop-list, shutdowns of separate word classes, not informative for a specific case is possible (nouns, verbs, numerals, etc.). The contextual report representing the literal maintenance of key fragments of a talk on the revealed subject is also available. Search of a thematic cluster in the expert mode providing creation by the user of own subject with a target set of the key word characterizing, according to the expert, a topic of conversation is possible.

Speech analytics for security services

The system of automatic detection of subjects is capable:

• without participation of the expert to create selections of phonograms on the subjects requiring constant and fixed control from specialists of security services
• in the automatic mode to reveal clusters of telephone conversations in which there is, for example, a discussion of matters of confidence and the commercial interests of the parties, transfer of personal data or important technology features
• reveal from a many thousands array of phone calls of the enterprise negotiations of employees on household subjects and discard them by consideration of incidents of production character.

System operation of a thematic clustering does not require any training of engines of speech recognition, selection of dictionaries of controlled lexicon or other technology procedures. A system will start the analysis of the voice information at once as soon as gets access to phonograms of calls.

Efficiency of implementation

• Repeated reduction of search time of target content in arrays of the unstructured voice information
• Increase in efficiency of actions by means of total automatic monitoring of a talk for target subjects
• Prevention and warning of incidents due to identification of a talk on new and uncharacteristic subjects
• Performance improvement of processing of the voice information identification of clusters of a talk on unproductive subjects