CityFALCON Voice Assistants

The idea was to give users an opportunity to use CityFALCON via the voice. The main goal was to set up voice search and allow CityFALCON’s customers to voice-search the stories by topic, publication, and time interval; read, like and bookmark them; create and manage watchlists, and many other things. At the same time, this integration was supposed to be easy to manage and support.

Connection of 3 voice assistants to the one site

We needed to connect 3 different assistants: each of them required different and complex configuration and training. To make it real, we simplified the workflow on the online voice assistant side and made it send the original user’s request as plain text.

Specifying the users’ needs

How do you make an AI assistant? To handle users’ requests, we need to understand what they asked for. Since we rejected the idea of using built-in assistants’ training systems, we were in need of replacement. Here is where LUIS comes in. It’s a machine learning-based service that is used for building natural language into apps, bots, or IoT devices. We used it as a single gateway to handle all requests after we added voice recognition to website.

Teaching the program

Voice assistant technology requires teaching. LUIS can parse plain text and detect user’s “intent” (what the user actually asked for) and “entities” (different keywords that identify search parameters, dates, topics, and so on). We configured more than 15 intents and taught LUIS to recognize hundreds of thousands of entities for our AI voice assistant.

Handling the users’ requests

After finding out what a user wants, we need to make sure that we can do this. To make it possible, we built a complex intent handling structure for our voice recognition software, where each intent has its own processing logic. As a result, each user request may perform some actions or return various kinds of information.

2019