Adventures in Speech Analytics – An Introduction

A lot of people in the contact center scene have heard about Speech Analytics (SA) but do not know the purpose of the technology. Researching the topic can be overwhelming so let’s start with the basics.

What is Speech Analytics?

Speech Analytics is the technique of taking calls that have been recorded to acquire data. It is to call centers what Google is to the internet. You can search for keywords or phrases to identify specific topics or behaviors during a customer-agent interaction.

What can I use Speech Analytics for?

You can use SA for a lot of things. Say your company requires sales agents to read to the customer a disclaimer about E911 (Enhanced 911) at point of sale- you need to know the percentage of agent compliance in your sales queue. You could spend an eternity going through calls randomly and hope to find what you are looking for or you can use SA technology to search for the specific phrase you are looking for like “E911 will not function until power is restored” and examine the data for accuracy.

You can also use SA technology to improve an existing process. For example one of the biggest issues in any size call center is First Call Resolution (FCR). You want to make sure that your agents are handling your customers concerns the first time around. The more customers have to call in results in lower customer satisfaction which could result in your customers ultimately leaving you for a competitor. SA can help you identify the hot topics that trigger repeat calls and help you develop processes to resolve them.

How does a Speech Analytics tool work? How does it know to find what I am looking for?

A SA tool typically is broken down into the following functional segments:

Speech Engine- This portion does the initial analysis of audio and will convert it into a file that contains a series of phonemes or transcribed text.
Indexing- This portion improves accuracy of the speech engine. It will try to make sense of the data from the speech engine and index it for additional analysis, ad hoc searching and queries. The tools will import/ingest data from the call center’s recording system(s).
Query Engine- This is where authorized end users are able to define their queries and results they want in the SA tool.
Search- Used to perform ad hoc search terms on indexed audio files. Operates similar to a google search.
Reporting Dashboard- This is where you can go to organize system data in a customizable and easy to visually digest manner. In this area you are able to drill down and filter results to your needs.
There are 2 primary types of speech engines:

Phonetic Speech Engine- the English Language has 40 phonemes. A phonetic based SA tool will search your keywords and phrases broken down in phonemes in order to return your search results faster. Example: “an-l-it-iks” or “Foh-net-ik”. A benefit to a phonetic based SA tool is that it is not limited to a pre-defined dictionary.
Text to speech (also known as LVCSR) Engine – A text to speech tool will convert speech into text using a dictionary. It will then locate the occurrences of query words in the transcribed call. A benefit of this type of SA tool is that it creates a transcript of the customer-agent interaction.
What is a Query?

Queries are an absolute must to have in your SA tool in order to get the best results. A query is a prebuilt search that has been created within the SA tool to help an end user best refine their ad hoc searches. An organization will typically have some resources set aside who specialize in creating specific queries that are of significant value to the organization. For example a telecommunications company may want to have queries built around the core lines of business they offer in addition to having queries built for specific call drivers like payment arrangements or general inquiries. Typically queries should validate at 90% accuracy to be considered functional as it will reduce the amount of false positives and missed hits.

What is an ad hoc search?

An ad hoc search is a custom search performed within the SA tool. This is the exciting aspect of speech analytics as your options for what you can search for are relatively limitless. Ad hoc searches are the core of any SA program because they allow your organization to deep dive into the heart of your customer-agent interactions. Generally the best way to ad hoc search is when they are run concurrently with an existing relevant query within the system to vet out the best results. Keep in mind that you will need to validate the accuracy of your results to ensure that you are gathering the most relevant results for your ad hoc search. Further elaboration on the ad hoc search process will be discussed in the next article.

As you can see there is a lot of power that a good speech analytics program has to offer. There is an increasing need for contact centers to deliver better customer experiences along with reducing cost and improving agent performance. People are now starting to understand that a customer interaction is a revenue generating asset that contains a vast amount of insight data that needs to be tapped into, understood and acted on.

See you soon in part 2 of Adventures in Speech Analytics when we take a closer look at queries and the ad hoc search process and strategies within a speech analytics tool.

Written by: Diana Aviles, Operations Manager, Speech Analytics at BPA Quality