There are several components to ALEX’s speech recognition and response, all of which happen in real time. The first component is using Google Cloud Speech API to listen to what you’re saying and transcribing it to ALEX’s simulation Log. Then, ALEX determines if you’re asking him a question or making a statement. From there, ALEX has a unique algorithm that directs him to the appropriate answer. ALEX then vocalizes/synthesizes his response based on the selected scenario.
What type of information does ALEX use to communicate with Google Cloud Speech API?
A short (few seconds, but never more than 60 secs) audio recording snippet is sent to Google Cloud Speech API whenever speech recognition is requested.
Are voice recordings kept, reviewable and erasable?
No voice recognition snippet recordings are saved or stored neither at our servers nor inside ALEX.
Is ALEX always listening like other voice assistants?
No. The short audio snippet is only captured and transferred to Google when conversation has been activated by touching the shoulder of ALEX and as indicated by the "bebeeep" sound. Listening stops when touching the shoulder again and playing the "bebuup" sound.
Is the data/information encrypted?
All data transferred to any cloud services are always encrypted using HTTPS. Additionally, any ALEX data and control is only available using encrypted HTTPS connections.
Is the speech data used to improve the AI?
We only use the written transcript to improve our response selection algorithm and not the audio.