The fewer words the better?
In everyday life, we try to use accurate, competent speech. And the longer and more voluminous (filled with meaning) sentences we make up, the more well-read and intelligent we seem to the interlocutor. However, the voice assistant is not such an interlocutor as a living person. If the sentences are long and have a lot of lexical turns there is a high probability that it will understand first part of the sentence or do exactly the opposite, rather than what you said. Though it might recognize your command. Therefore, it is better to use simple sentences. There is also an occasion when a request is received in two questions at once, and the column, in turn, may answer only one of them, and the other - either remember or do not take into account. In such cases, the voice assistant may become confused and lose the sense of the dialogue. "How could I address you? Be kind, please"
It is known that some dialogue assistants have a personality. If we are talking about a smart speaker from Amazon, it has integrated Alexa - voice assistant, which is: "I'm Alexa! I'm 6 years old. I can provide you with information, music, news, etc. And I also like to make jokes, such as telling jokes about unicorns!" "Personality" includes almost all the qualities that a living being has. Thanks to this, during the dialogue, you can find out the bot's biography, character, it's purpose, musical preferences, and much more. If you ask Alice from Yandex to tell you something about itself, it even plays a joke: "I was born in Siberia with a stubborn soul. And as I was loved, I grew up big. Joke". According to our teams statistics, 8% of all the dialogues users want to learn "who" the bot is.
Therefore, we are also working on the personality of our Dream Bot. It has the ability to prove people that it is their friend and assistant, but it is still not quite aware of the world around it and people in general. So, it capability is currently limited, but the bot is constantly expanding the world awareness and becoming better. And this is only part of its credo. How do you imagine a voice assistant? Should it be a "boy" or a "girl"? Is it important for you? Emotions rule the world
Do voice assistants take into account the user's emotions? Сurrently, some companies are working on creating a similar system. By analyzing the user's facial expression and voice intonation, the system can determine and evaluate the current mood. Speaking about emotions...The voice assistant show no emotions. It can say how happy it is and it says it in a flat, monotonous voice. But it's not that bad. There are voice assistants which have some phrases with intonation. For example, if you talk to Alexa in a whisper, it answers you in a whisper too. Try it! It's funny!
Voice assistant features
It should be understood that there are various smart speakers and they solve certain goals. Some will make purchases for you, others will make a playlist based on your taste preferences, and others can respond to certain commands or talk about some topic.
As part of the competition, Dream team focus on communicating with the user during a long period of time, so it is social. Such conversation should be meaningful, interesting, and cover current events and popular topics. To implement the structure, we use the DeepPavlov Agent
. It helps bot developers assemble multiple NLP models into a single pipeline. Because modern chatbots usually have to solve different tasks at the same time. Personalization
Due to personalization, there is a capability to adjust the function to fit your preferences. One of the cool features is the ability of the smart speaker to recognize the voice. For example, if your guest asks you to include a favorite playlist or put a like on a track, the smart speker will refuse to do that and issue a message: "You are not my host!". But as part of the competition Alexa Prize Social bot Grand Challenge 3, we are not working on voice recognition. Non-native speakers
How is it!? Since our Dream team participates in competitions from Amazon, the smart speaker understands only one language - English. And you know there are different dialects in any language. When interpreting, for example, Russian into English, the user may encounter different complications. Sometimes the "operator "can turn into an" Emperor" or vice versa. It is not easy for the residents of India, Spain, and Italy, where the native language imposes its own style of pronunciation. Therefore, you should not forget about the accent when developing a voice assistant.