Speech Analytics
Speech analytics is not a difficult solution to bring on stream. On the basis that an existing call recording system is in place, interfacing the two systems to feed the call recording archives into the speech analytics system can be rapidly achieved.
Of greater technical concern is the decision around which speech analytics technology to adopt. Speech to text conversion is the most widely promoted solution and one currently preferred by the call recording vendors. For those that have already invested heavily in their performance optimisation portfolios, there is much to recommend the pre integration benefits of such an approach to speech analytics.
However a robust and credible alternative is to be found in solutions that work on the basis of phonetic indexing and analysis, although for the sake of accuracy all approaches to speech analytics have some phonetic component in their processing.
Indeed some vendors are augmenting the post processing phase of their solutions with hardware based phonetics boards to gain some of the benefits that phonetics brings to the analysis of unstructured conversational speech that typified customer service interactions. But these are still dependant on a pre determined lexicon of words and phrases to function.
Solutions that are heavily geared towards the phonetics approach can process voice files at extraordinary speeds and do not suffer problems of word recognition such as real names or slang since they are not dependent on the completeness of dictionaries which are used as reference in speech to text solutions. This is a problem that can become even more exaggerated when less common terms or phrases are being used in searches which therefore impacts the accuracy of the search.
Another point of consideration is the fitness of the solution in the context of call centre environments which have typically noisy backgrounds, speakers with differing accents and dialects and significant percentages of non-native speakers in their customer mix. The native ability of each type of solution to work in this real world environment is crucial in delivering accurate results.
In summary the issues for consideration are ones of processing speed and accuracy of results. In relation to this last point, dictionary training, implementation, maintenance and the overhead this creates plus the initial phase of dictionary training which can take anything from one to three months, needs careful consideration.
See alternative business needs description
Datapoint provides the following technologies and services for Speech Analytics.