The current boom in artificial intelligence could not do without investing our homes: not content to be omnipresent in the heart of our smartphones; personal voice assistants are also trying to find a place in our homes today.
The small sector of the connected home is being shaken up by the arrival of these new types of devices, which are sometimes favoured by Google Assistant, sometimes by Alexa and sometimes by Siri or Microsoft. These assistants, we can also describe them as intelligent speakers boosted with artificial intelligence, intend to occupy a prominent place in homes.
Key points to consider
Since smart speakers are intended to be permanently or almost permanently accessible, it should be possible to consider placing them in any room of the house (living room, kitchen, bathroom, etc.) and even why not outside, since some have batteries. Some attention will, therefore, be paid to the robustness of the product, in particular by monitoring any IP certification (water or dust resistance).
If the speaker is designed for music listening, we of course subject it to the same versatility requirements as a “non-intelligent” speaker. It is, therefore, necessary to check whether it offers complete connectivity, as well as exhaustive and easy-to-use controls.
Like all loudspeakers, “intelligent” models have to offer quality sound reproduction. It is to ensure not only that they can provide a pleasant music listening experience, but also to reproduce the voice of the vocal assistant in a clear, natural and perfectly intelligible way.
It is a good thing that the user can understand the assistant, but it is apparently no less important that the assistant can recognize the user! Not all intelligent speakers are equal in this respect, succeeding in picking up voice commands from a greater or lesser distance, being more or less sensitive to surrounding noise and benefiting from a more or less effective understanding algorithm.
We test audio performance in an utterly classic way: our acoustic measurements mainly include frequency response, harmonic distortion, power, impulse response, etc. Speech recognition quality is measured by sending several voice commands to the speaker at a calibrated sound level, pronounced by a sample of people with different voice timbres.
We then check how far the control is perceived, with what background noise level, etc. Outside the lab, all these products are tested in a “connected” environment to observe the possible interactions with the various equipment in the house.