If you are looking to implement speech analytics software at your contact center, you have probably noticed that there are two main technologies out there. The first is transcription, also known as Large Vocabulary Continuous Speech Recognition, or simply LVCSR. Transcription aims to convert the entire audio conversation into text, based on dictionary-recognized words. The second technology is usually referred to as phonetic indexing and search. Rather than transcribing the entire conversation, it converts it into a string of phonemes, the basic units of speech. It then looks for only a predefined list of words.
Different solutions on the market typically revolve around one of these technologies. So which one is better?
Phonetic indexing is much faster than transcription since it does not need to scan tens of thousands of dictionary words. As a result, not only can you achieve quicker business insights, you can also analyze higher volumes of calls using the same amount of servers. In addition, phonetic-based analysis is the only practical way to carry out real-time speech analytics in order to probe calls as they happen and trigger agent guidance or alerts when issues are identified.
Phonetic indexing has also been shown to offer greater accuracy.
Sounds like a winner, right? Only it lacks the ability to conduct deep data mining for root-cause analysis. So you’ll need to know what you are looking for. Most of the time this won’t be a problem, but once in a while a new issue will come up and a phonetic indexing solution will not identify it for you, leaving you behind the curve.
Transcription on the other hand allows the system to identify issues without first pre-defining them. It can tell you why customers are calling, show you the repeating issues across various calls, and conduct root-cause analysis to identify the drivers of such issues. So is this technology the winner? Not so fast.
Transcription is much slower than phonetic indexing, requiring many more servers, and when dealing with high call volumes you’ll often have to settle for analyzing a sample of calls. In addition, your analysis will be dictionary dependant and won’t recognize words such as product or brand names.
So no one wins by a knockout. Transcription will one day be fast and accurate enough to render phonetic indexing unnecessary. But that day is not today. Does that mean you are doomed to pick sides? Not necessarily.
Why not combine phonetic indexing and transcription into a single solution and enjoy the best of both worlds? You can analyze 100% of calls with fast phonetic indexing, categorizing them into call types, viewing the trends, and identifying problems, for example, an increase in customer dissatisfaction or a spike in calls about billing. You can then transcribe call categories of interest to conduct root-cause analysis. This way, you are transcribing the calls that matter rather than wasting time and resources on those that don’t.
For a more detailed review of hybrid speech analytics software technology, see this white paper.