Voice bots have already become valuable employees who can provide better service for
One of the options for solving this problem ispre-recorded script for various situations, including a conversation with an angry interlocutor. It allows the voice bot to adapt to the situation and provide relevant assistance. This can be called programmed empathy, the main condition of which is the ability to correctly recognize a person's emotions and respond to them. And the current level of technology development significantly increases the EQ (from the English Emotional Quotient, "emotional intelligence" - "High-tech") of voice bots. For example, TWIN bots correctly identify basic emotions in 96% of cases: positive, negative or neutral.
But creating empathetic bots that cansimulating EQ thanks to pre-recorded scripts requires large computing power, which means their cost automatically increases. This is not always profitable for business. This is why the robot often switches the conversation to a live employee as soon as the dialogue requires empathic communication.
How voice software recognizes a person's mood
Voice emotion detection is a microservicegradually learning by example. To do this, they collect many dialogues between the bot and the client and analyze their quality, assessing how correctly the robot recognized the emotion. Errors are corrected and replaced with the correct version in the system.
However, this cannot be done without a person.The analysis and labeling of emotions is carried out by linguists who clearly define the shades of speech and reactions: positive, negative or basic. They listen to a lot of dialogues, from which they select episodes with pronounced emotional overtones, categorize them and enter them into the appropriate database. This creates a set of datasets or models that form the basis of the bot’s emotional intelligence. It is very important that the marking is done by a professional, since even a small error can completely “kill” the entire dataset.
After the labeled sample modelThe data is ready, the neural network training phase begins. At this stage, in addition to the quality of the datasets, the best ML engineers (machine learning specialists - Hi-Tech) are needed. In addition, it is worth choosing the technology stack carefully. This is the only way the system will work perfectly, recognizing the emotional coloring of almost any speech, with the exception of whispers.
The main difficulties in recognizing emotions:
- Variety of emotional shades- Even a person can find it difficult to distinguish between concern and irritation. That is why they are currently distinguishing between three basic emotions: this allows us to reduce the error.
- Expression of emotions— the robot often relies onto verbal formulations. However, this approach is not always effective. For example, obscene language in itself does not necessarily indicate negativity. Therefore, TWIN divided two emotion detection microservices: one is trained on voice and is used in calls, the other is trained on texts and is used in chats.
In what areas do bots need to be able to distinguish emotions?
Service support
Service support often involvessituations when the client needs to be quickly transferred from the robot to the operator, especially if the person has a negative attitude. Companies in the transportation industry are often faced with a typical call: an angry user calls a contact center who did not receive an order on time. Having received the call, the robot hears a lot of criticism against the company, but cannot adequately respond. We learned a lesson from such situations: now the call is automatically transferred to the operator, as soon as the system identifies negative emotions.
Emotional intelligence becomes part of the servicein e-commerce. For example, Amazon taught its voice assistant Alexa to recognize emotions by intonation. The function is likely to help improve the communication quality of the smart assistant and users. And Emotibot offers a voice assistant for post-sale e-commerce services. The bot adjusts the responses to the mood of the interlocutor to prevent the escalation of complaints.
Analytics
Call centers are also starting to experiment withemotion recognition technology in order to evaluate the quality of service. For example, the system can analyze the operator’s speech for the occurrence of negative connotations of speech: this controls the communication of employees with clients and prevents regular violations of communication ethics.
Voice technology with emotion recognitionThey also help to analyze the dynamics of customer sentiment: for example, to determine whether customers in general have become more irritated or, conversely, have found more brand loyalty. This gives an understanding of whether the automation of the call center is going well or if some processes need to be improved. For example, a similar technology was tested by Rosbank. As a result of the analytics of the dialogue between the bot and the client, the manager receives a statistical report for each call, which reflects such data as the dynamics of the consumer satisfaction index and a comparative indicator of service efficiency.
Sales
Emotional state directly affectsbuyer's willingness to make a purchase. It is easier for a person to sell something when he is in a good mood. Research confirms this: for example, a Belarusian experiment demonstrated a direct connection between positive emotions and the willingness to communicate with a sales manager, as well as overall brand loyalty.
Recognizing emotions can increase income frombots that handle lead processing. So, for example, relying on this marker, a salesman robot is able to immediately transfer a call to a “live” employee so as not to miss a “hot” lead. Probably, in the future this tool will be used by the majority of companies where bots are engaged in calling a cold or warm base.
The development industry has already proposedonline retailers a guide bot that can act as a digital consultant and build a dialogue with the user based on his character (the bot determines him by his emotions). However, such solutions have only just entered the stage of study and testing by the market, so they have not yet received mass distribution.
Should the robot recognize the sex of the interlocutor
Clients want to see an individual approach toservice in any industry. In a SmarterHQ survey, two-thirds of consumers admitted that they only engage in conversations when they receive personalized messages. This dictates the need to continue to work on making voice bots smarter, segmenting interlocutors based on various criteria, the most obvious of which is gender.
Gender recognition allows you to build speech withtaking into account this criterion, and also reduce the percentage of errors when dialing a subscriber. For example, if Natalia is listed in the database and a male interlocutor picked up the phone, the bot will be able to instantly react and say: "Can I talk to Natalia?" instead of "Hello, Natalia".
It's the same with calls.Let's simulate the situation: the robot is recruiting a client whose name in the database is Natalya. Accordingly, he switches to the mode of communication with a woman. But suddenly Natalya’s husband or brother picks up the phone and says: “Sorry, I’m not Natalya.” If the robot does not determine the gender of the interlocutor by voice, a funny situation may arise: the client will prove that he is not Natalya, and the robot will continue to communicate with Natalya, and the conversation will most likely end very quickly.
Of course, you can completely abandon names andaddress everyone in a unified manner, because in approximately 3% of cases the phone is answered by a person who was not registered during registration. But for retailers, for example, it will cost a few loyalty points. Just imagine: you ordered a product from an online store, and before sending it, a bot calls and asks if it will be convenient to receive your order in the near future. A nameless address will make the conversation cold, while a simple “Hello, Sergey!” will soften it a lot. It’s very useful to remember Dale Carnegie’s catchphrase: “A person’s name is the sweetest and most important sound for him in any language.” This is confirmed by scientists Dennis P. Carmody and Michael Lewis, who proved that our brain goes into ecstasy when we hear our name, which instantly focuses attention on the sound.
So to avoid awkward situations,while not sacrificing consumer loyalty, it is necessary to distinguish between men and women by voice. Our bots, for example, correctly determine the sex of a person in 98% of cases. In the West, a similar function is equipped with My Voice AI, which can also determine the approximate age and emotional state.
Robots are just beginning to explore the depthshuman speech and distinguish the client's mood or gender. Technology is likely to advance rapidly in the future. Gartner predicts 10% of all devices will be equipped with emotional diagnostics in two years. This will inevitably stimulate the development of the market for systems for analytics and emotion recognition: according to data from MarketsAndMarkets, it will grow from $ 21.6 billion in 2019 to $ 56 billion by 2024 with an average annual growth rate of 21%.
See also:
Astronomers have found an exoplanet with a strange orbit
The Doomsday glacier turned out to be more dangerous than scientists thought. We tell the main thing
Earth is in the process of the sixth extinction. How will this affect humanity