1. The human speech inputAs said, voice assistants are available either integrated in a device, through a stand-alone speaker or through an app on your mobile phone. But regardless of how you engage with the voice assistant, it still all starts with you and the input you provide! Let's say you want to go to the movies next weekend with a friend but you're not sure when or where a certain movie is playing. You can ask your voice assistant to help you gather this information by asking: "At what time next Friday is Alice in Wonderland playing in a nearby cinema?". This question is called an ‘utterance’; it is the input that you pronounce.
2. Conversion of speech input to text inputThe voice assistant will try to describe what it heard and convert to text what it understood. Hopefully, it understood you correctly and the description is literally the utterance you pronounced. Sometimes it doesn't understand you correctly, but the assistant will always try to convert what it heard to text. It is likely the description is still a correct sentence or at least using correct words combined into a sentence. The natural language processor (NLP) used in the voice assistant will always convert something into words that are part of a specific language dictionary. However, the result, can be a description that does not reflect your utterance correctly.
3. Handling the text inputOnce there is a description of the utterance, the voice assistant still needs to understand the description. A voice assistant can listen to a lot of different requests, questions or remarks on a lot of different topics. Somehow, it needs to know how to handle your request. One of the ways to give structure to a large set of topics, is through categorization of the input. This categorization is called 'intent mapping': it tries to understand the intent of the textual input and map that to a specific target. An important part of developing a voice assistant is training this intent on many variations of input. In our example we could map our question ("At what time next Friday is Alice in Wonderland playing in a nearby cinema?") to an intent SearchMovieTimes.
But that’s not all it does. Next to mapping the input to the right intent, it will also try to catch pieces of useful information from that given input. These small pieces of information are called ‘entities’. In our example this could be the date/time (next Friday), the name of the movie (Alice in Wonderland) and perhaps the name of the cinema (which was not present in our example ques-tion) or the location of the cinema (nearby).
4. Processing the requestAfter it has mapped the intent of the request, the next step can take place. Any proper voice assistant is programmed to gather all relevant information before it provides a response. It could access anything from restricted internal company systems to publicly available data sources.
In our cinema request, if it does not have all the required information, it could respond by asking additional questions to gather missing information. For instance, it could ask for a more specific date, or how many people are joining. It could even ask permission to access your current location, to pinpoint how it should interpret 'nearby'.
5. Response formulationOnce the required information is gathered and the relevant answer to the request is searched for, it can prepare the response. The voice assistant can formulate a textual response following the linguistic patterns of the specific language. The response should provide a best possible answer to the request and of course, should do so in a friendly manner. In our example, it would use the different entities to search for the movie in the available Movie API’s that are connected.
6. Text to synthetic speech outputFinally, it will pronounce the formulated response in audible output. Most voice assistants could be connected to use one of many different synthetic voices, for instance a male or female voice. The answer in our example could be: "Sure! The movie Alice in Wonderland is playing next Friday, the 26th of April at half past seven at Kinepolis. Do you want me to make a reservation?"
At this point, the conversation could either end or continue by guiding you through the ticket reservation process.
How to prepare your organization for the voice channel?While the given example is pretty straightforward and quite linear, it is possible to create many different voice assistants for many different purposes. So, if you are thinking about creating a voice assistant for your company, no matter how simple or complex a use case you are considering, there are a few things to consider.
1. Understand the user needsOften companies reach out because they don't want to miss out on the voice hype. Whether someone from the marketing team found their new gimmick or someone from IT is really excited about the new voice speaker he just bought, the underlying reason for jumping in is unclear. Having a voice assistant has become more important than a clear purpose of the voice assistant.
The risk of creation without purpose is long term. Exploring the potential of the voice channel should always be seen as part of a larger conversational or even omnichannel strategy. It's just another channel to communicate with your users. So always research and understand the needs of the user before thinking about launching a use case. Use this user input to create a concept that fits their needs. And make sure the voice assistant enriches your customer journey.
2. Invest time to train and maintain your voice assistantA voice assistant is above all a programmed state of reality. While it can automate certain tasks, the success and value of the experience is determined by how well we capture the intent of the user and the quality and combination of data sources that power the experience.
Besides dialog design, the training of the voice assistant is important. It’s an ongoing activity once the assistant is in production and used by your customers. It is good practice that you (or the team of your digital partner) spend ample time on training the voice engine on all the different intents and entities. For instance, UX researchers and dialogue writers could help you in discovering how users will interact with your voice assistant.
Don’t forget to think about handling those scenarios where a customer cannot be helped automatically by the voice assistant or where he is not being understood properly. Human take over remains crucial to improve a bot service. Include your customer service team and prepare them for these additional questions. So, in case a customer does get frustrated, make sure the bot can switch to a human being.
3. Use the value of analyticsOnce you are satisfied with the first test results of your newly developed voice assistant, you are ready to release it. Often when releasing the first iteration of a new service, it is recommended to soft launch first to make sure any overlooked issues can be identified and resolved quickly. When you are confident it is ready for the broader public, activate your marketing and communication campaign. Once people are using your new voice assistant it is important you understand how they are using it. Gathering insights to understand how users are actually using your voice assistant is extremely valuable. The data can tell you how people perceive it, where they are not un-derstood correctly and possibly get a bit frustrated, but it can also provide you with new insights into desirable functionality. Prepare your business to gather these data points and make sure the team has the ability and agility to adjust, improve or train the voice assistant to cater to the needs of the user.
Need help?In order to understand the possible use for your business, I suggest you talk about it with peers, join an event like Open Voice or just give it a try yourself. At Mirabeau - A Cognizant Digital Business, we have already created many different experiences for the voice channel for our clients. We can create valuable voice solutions that help you increase your digital footprint. We offer a voice design sprint format which generates valuable ideas and delivers working concepts that can be tested, to inform next steps on how to add value to your organization, for instance by providing the ability to schedule an appointment, by providing automation of your service desk, offer help in determining your next travel destination, get status updates of your purchases or by providing guidance in putting together your DIY piece of furniture just to name a few examples.
Do you want to know more about Conversational Interfaces or Voice assistants? Please contact Eric Vanderfeesten