Alexa, who am I? Amazon’s Alexa learns to understand who’s speaking with her

Amazon’s Alexa often misunderstands what you and your household members are asking her. But, she has a trick to improve her understanding: she can learn to recognize who’s speaking.

How can you teach her who’s speaking? You can repeat a couple of phrases that she asks you to pronounce, based on which she will create your voice profile.

That way, every time you say something to Alexa, Neural Networks will compare your voice to the profiles created, and understand with whom she is speaking.

The trick that Alexa uses is that the voice profile is created once she asks you to speak, so that the noise is reduced and little preprocessing is required.

How is the profile actually created? Artificial Neural Networks from the voice receive signals, that are then translated into vectors. Those vectors are encoded using Autoencoders technique that compresses the voice profile into a form that can later on be decoded without huge information loss — it works in the same way as image compression. Once you speak again to Alexa, the new voice vector is encoded with the same Neural Network, and the decoded output is compared to the voice profile created.

After creating your voice profile, you can ask Alexa to help you out by calling someone, assisting you with your shopping, or playing a music for you, all while she personalizes your experience.

Don’t forget to sign in with other household members’ accounts as well, in order to let Alexa connect their voice profiles as well, and recognize who’s speaking to her.

To make sure Alexa knows you, you can always ask her “Alexa, who am I?”.