Speech applications that improve caller experience

The Importance of Effective Tools

Great speech applications just don't happen by accident. Great speech applications are the result of a rigorous and systematic development approach supported by an extensive set of development tools designed specifically to achieve the best possible application performance.

While most of the industry was focusing on call flow implementation tools, we were focusing on building tools — grammar tools, tuning tools, dictionary tools, simulation tools, testing tools, etc. — that enable us to deliver rock-solid applications with unparalleled user experience and automation rate. The results speak for themselves.

Grammar Development

Because grammar development is such a critical component of a speech application development project, we have built the most sophisticated grammar development environment available anywhere: The NuGram Platform.

suite-of-development-tools.png

Call Analysis

The ability to perform in-depth call analysis is critical in order to rapidly identify and fix problems with a speech application. Nu Echo has a full suite of Eclipse tools to transcribe calls, analyze them, produce reports, etc.

Speech Application Tuning — Atelier

Nu Echo's tuning activities are supported by Atelier, the most powerful integrated tuning workbench available anywhere. Atelier's numerous capabilities include:

  • A Batch Speech Recognition and Analysis Environment — Tuning a speech application requires the ability to iteratively perform off-line speech recognition experiments with the grammars — static or dynamic — used by an application and analyze the results in order to improve speech recognition accuracy, set confidence thresholds, test post-processing algorithms, etc.
  • A Confidence Score Training Framework — Confidence scores are extremely important for implementing effective speech dialogs, for instance to decide when a confirmation or a reprompt is required. Unfortunately, confidence scores produced by commercial speech recognition engines are designed to produce acceptable results for all grammars, which means that they cannot be expected to give optimal results for any grammar. When field data is available for a given grammar context, it is possible to train much better confidence scores and therefore significantly enhance an application's performance.
  • A dialog simulation environment — Dialog level performance metrics can only be estimated through complete dialogs. That's why we have implemented a dialog simulator that enables us to simulate the impact of certain application parameters (e.g., confidence thresholds) on dialog level performance metrics.

Automated application testing

We believe it's not the role of users to find and report bugs in a speech application. It's therefore critical that speech applications be thoroughly tested before being deployed. We also believe that, although human tests are absolutely necessary, anything that can be automatically tested should be. We have found time and again that automated tests make it possible to easily and rapidly find and correct problems, usually as soon as they have been introduced. Undoubtedly, without extensive automated testing, many such problems will find their way into the deployed application, therefore not only annoying callers, but requiring unnecessary bug fixing and testing cycles.

Multilingual Phonetic Pronunciation Management

High performance speech applications require accurate phonetic pronunciations. This is as true for speech recognition — where incorrect pronunciations may result in speech recognition errors — as it is for text-to-speech — where incorrect pronunciations are likely to result in a call automation failure. Getting accurate pronunciations, however, is not trivial, especially for large vocabulary applications. To address this challenge, Nu Echo has developed a sophisticated multilingual pronunciation framework that enables us to efficiently produce accurate pronunciations for large specialized vocabularies.