Layer 2 technology provides complete user-experience support for IVR systems that require speech recognition. Users of SPT Layer 2 technology will experience the most effective mixed-mode dialogues available from current, state-of-the-art spoken user interface research.
Features Layer 2 features include all Layer 1 Foundation features, plus:
ShadowPrompt® Scripting
Callers may choose to use speech or touch-tone at any point in the dialogue, and in fact cannot know in advance which modality they will need. Layer 2 scripting methods give uses the freedom to switch modes at any time. In the SPT model, an IVR user interface is a dual-mode interface – both speech and touchtone are available concurrently. But one of the modes is a “foreground” mode – currently the preferred mode, and the other is in the “background” (or the “shadow”).
A ShadowPrompt announcement delivers the foreground prompt followed by the secondary mode prompt as needed. Delivery is carefully designed to interact with the user’s memory and context, encouraging simple and accurate decision-making. Timing, sound quality, and surrounding pauses are all considered.
The design of these prompts is well-researched and the subject of SPT patents. In other words, they work. Users report that they move seamlessly through the IVR menus with little conscious thought about whether they are relying on touch-tone or speech to get the job done. In fact, we have seen many users in usability tests that simply don’t remember which modality they use. This is because they were thinking about their task, not about the user interface.
Point and Speak Selection
Layer 2 supports list selection for both speech and touch-tone. For the most accurate speech recognition, users repeat the list item just heard. The list selection approach compares the n-best list from the recognizer against list elements recently spoken by the IVR to resolve ambiguity. This drives up first-time accuracy, and reduces errors associated with low confidence rejections.
Novice users may stumble on their speech or present out-of-grammar utterances when recognizing a list element that they want. The point-and-speak feature treats the resulting false acceptance or rejected “no-match” as an event in time that is “pointing” at the list element they want. Rather than engage in error recovery dialogues, the algorithm gracefully interacts with the user about the list element that was just presented when user speech began.
Just-In-Time Navigation
Navigation commands often clutter IVR menu lists to the point of exhausting caller short-term memory. Moreover, users can rarely remember navigation commands from one dialogue state to the next. Layer 2 Just-In-Time navigation interjects ShadowPrompt tips where and when appropriate reducing cognitive load on the user while maximizing IVR performance.
Foolproof Yes-No Questions
One would think that the simple execution of a yes-no microinteraction would be one of the simplest of all directed dialogue behaviors. But in fact, yes-no questions continue to be fragile in most IVR systems. The grammars are too big and therefore sensitive to false acceptance, and so they fail in noisy environments and in the face of chatty users.
Layer 2 yes-no questions are stable and transparent and dramatically dampen error amplification.