The first voice-activated assistants appeared on smartphones in 2011. Initially confined to smartphones, this software application based on natural language speech recognition and the reproduction of information by voice synthesis is now being deployed in other universes: connected speakers, audio headsets, vehicle interiors, and even other everyday objects, which are thus seeing their interaction possibilities renewed.
Coupled with artificial intelligence, this technology, which has been around for a long time, is now coming of age. Voice assistants understand on average 90% of the requests they receive. According to the specialized firm comScore, half of all Internet searches could be by voice by 2020. To be massively adopted, voice recognition will still have to overcome some difficulties.
Judging by a whole series of surveys and polls, voice control would have massively entered into usage. Julia Velkovska and Moustafa Zouinar, who have been conducting surveys on the actual uses of voice assistants since 2015, observe, however. "a real gap between the promotional speeches that praise the conversational capabilities of assistants and the reality of uses".
One French person out of two uses a voice assistant, 19% regularly
According to a survey conducted by YouGov, 56% of the French population claim to own a device with a voice assistant: 74% of 18-35 year olds and 45% of those aged 55 and over.19% of them say they regularly use a voice assistant: 8% daily and 11% "often".
- 42% of users (regular or occasional) would trust their voice assistant to do research on the Internet
- 30% would use their voice assistant to find an address or directions
- 28% would use their voice assistant to send a message or make a call, 17% to ask for the weather and 16% to listen to music
- 57% of voice assistant owners would say they are satisfied with the answers provided by their voice assistant and 11% of them are even very satisfied.
- 19% of users use them "often", 42% "occasionally", 34% "more rarely" even than rarely.
- When they do, it is for simple Internet queries (86%), requests for weather information (69%), playing music (58%), writing a message (58%) or setting an alarm (50%).
- The more involving actions (shopping, booking a cab, home automation) still seem to require a learning phase and/or additional equipment.
- 51% of respondents believe they are not in control of the information collected by these voice assistants and nearly a third are concerned that they will have fewer choices of results offered.
To work, voice control requires "real work" on the part of users
Two researchers from the SENSE Laboratory (Sociology and Economics of Networks and Services) at Orange Labs, Julia Velkovska and Moustafa Zouinar, have undertaken, on the basis of video observations and interviews, to understand "how people appropriate or do not appropriate these new systems, the way they interact with them, the meaning they give to these practices, or even the place they are ready to give them in their home, in their everyday life . They observe " a real gap between the promotional speeches that praise the conversational capacities of assistants and the reality of their use. In practice, it is not as easy to talk to them, it often requires an effort on the part of users, whatever the system.According to the two researchers, speech recognition is not always effective. "Users sometimes have to repeat their statements several times to be understood, which can lead them in some cases to abandon the use of the system. In cases where they persevere, they engage in real "user work", related to the management of the interaction and its meaning. This work can take the form of a variety of actions such as rephrasing statements by shortening them or expanding them to provide clarification, approaching the object or speaking louder... The effectiveness of the adaptation work varies according to the person, according to their way of expressing themselves, according to their pronunciation, accent, speech rate, etc. While voice assistants are intended for collective and family use, not all members of the household are equal when it comes to the "user work" required for a successful interaction.
They observe, for example, that "some young children have a lot of trouble making themselves understood, which creates a feeling of collective frustration that also affects the parents and can lead to a situation of exclusion. But this can also concern adults who express themselves with a particular accent, for example. To cope with this situation, people try to help each other, for example by giving each other advice on how to speak to the assistant.
According to Moustafa Zouinar, " many questions remain open, notably those of knowing how users will appropriate them in the long term (will they become part of the daily life of households?), if voice assistants will transform (or not) the domestic activities of families and how, what types of relationship and attachment users will develop with them.
Challenges that remain to be overcome
The voice recognition technology on which voice-activated assistants are based is not new. IBM already developed the first tools of this type in the 1970s. Coupled with artificial intelligence, this technology is now reaching maturity. To be massively adopted, voice recognition will still have to overcome a few difficulties, notes the Digital Factory.The assistants do not understand the question or get the answer wrong: it is difficult for them to distinguish the human voice in a noisy environment or to recognize a sentence pronounced with an accent. Problems also arise when users use a specific register. Voice assistants "understand" standard sentences (for example, requests about the weather) but encounter difficulties when asked to use specialized vocabulary (business vocabulary, brand names, etc.).
Today, voice assistants understand an average of 90% of the requests they receive. The 95% mark could be reached by 2020. The designers of voice assistants are redoubling their efforts to achieve a final error rate of 1%.
Serious questions about privacy
Perceived as black boxes, with opaque functioning, voice assistants crystallize important questions concerning the protection of privacy. They are called upon to take a central place in exchanges and interactions, including the most intimate ones, and they have, moreover, the particularity of being permanently "listening" to detect the word that activates them. Several studies have shown that messages inaudible to the human ear can be sent to voice assistants, which not only hear them, but obey them. People have no way of knowing what is really being listened to, processed and stored. The CNIL has just devoted a very complete file to the privacy issues raised by these voice assistants.Références :
