Speech Analytics

When comparing speech analytics solutions it is essential to understand the subtle but important differences between direct phrase recognition and transcription and how the analysis is performed. Each method has an affect on the accuracy of the results that you’re looking for. Fonetic speech analytics analyzes 100% of voice call recordings using direct phrase recognition which is fast, highly scalable and provides you with the most accurate results.
Speech Analytics
Fonetic processing is based on direct phrase recognition which does not convert the audio into a transcription before the analytic processing – it analyzes the speech directly from the audio looking for specific phrases and strings of words that have been pre-defined as being important to your business. By analyzing the speech directly there is no loss of data, unlike any other solutions that convert the speech before analysis. Direct phrase recognition therefore, analyzes a more complete data set, which makes it the most accurate technology in the market. Independent tests show that direct phrase recognition is 5-9 times more accurate than transcription based solutions.
Transcription Processing
Most phonetic transcription is based on the assumption that linguistic sounds can be segmented into units that can be represented by symbols (letters).
Given that the pronunciation of words varies greatly among both native speakers and foreign speakers, not to mention the array of dialects and slang that must be taken into consideration, most transcription software will simply not transcribe a word that it doesn’t recognise.
Therefore, when the speech is transcribed, data is lost. Tests show that this can be as much as 50 – 60% which means the accuracy of the analysis is compromised by the loss of data.

Fonetic Speech Analytics Technology

Speech analytics is the process of analyzing recorded voice calls to gather information.
Using speech analytics you can identify business trends and issues providing you with facts to help you to manage your business more effectively. Such as:

  • Product, service or quality issues
  • Performance statistics
  • Customer churn and competitive threats
  • Compliance behavior
  • Trends
  • Upsell and cross sell statistics
  • Identify high value customers
  • Understand performance of marketing campaigns and customer reaction

In addition you’ll have visibility of:

  • Any topic or number of topics being discussed
  • The amount of speech versus non-speech (e.g. call hold time or periods of silence) and the location of speech in the call – so that it is easier and faster for you to find
Fonetic Speech Analytics

Speech Recognition

Speech recognition is the translation of spoken words into text. The Fonetic solutions have automatic speech recognition LVCSR (large-vocabulary continuous speech recognition) built-in.
The transcription provided is not used for analytics purposes, but as an aide for the end-user of the solution to be able to quickly reference a point in a conversation.

How Does it Work?

Indexing (pre-process)
A copy of the audio is stored as an index file. The index phase performs an analysis of the file, which enables the solution to optimize the speed of subsequent searches following this phase.
Data Structure
To be able to analyze voice data (which is unstructured) it needs structure according to the environment (types of conversations that are taking place) in which it is recorded.
Fonetic gives the data structure by identifying categories and topics according to the type of business that is using the analytics tool. This is the area where Fonetic has years of experience of building categories and topics that makes the Fonetic solution outperform competitor products.
For example, in banking, a category might be a financial instrument such as a commodity and the topics associated with that category (like a sub category) would be the types of commodities, such as coffee, cocoa, sugar, rapeseed oil, etc.
Metadata Enhancement
The data is further enhanced by the addition of metadata as it is important to analyze the data around the call to provide a clearer picture of events. Lots of metadata is available and might include information such as:

  • Telephone number called / calling number
  • Data and time of call
  • Call duration
  • Call handler’s name / ID

Once the data has structure, the Fonetic solutions enable you to extract key insights and information that is otherwise buried deep in the conversations that take place between your staff and your customers, or the people your staff are doing business with, such as in a financial trading environment.
Query Parsing
Whenever the end-user makes a request of the tool to find a word, phrase or provide an exception report the solution parses (analyses string of words) to find what the user is looking for.