How does a voice assistant turn sound into words?
Spoken words reach a microphone as changing air pressure, not as neatly separated text. A speech recognizer has to turn that signal into a likely sequence of words. The documented Listen, Attend and Spell model offers one example: an encoder processes sound features and a decoder generates characters. Recognizing a request is a separate problem from knowing how to carry it out.
Questions you could explore
- Why might two phrases sound alike but become clear once surrounding words are considered?
Start reading
- Google ResearchListen, Attend and Spell
The abstract describes a recognizer that converts sound features into a sequence of written characters.
Ready to try your ideas aloud?
Your topic carries over. Record up to 2½ minutes, depending on your plan.