Speech Recognition

What is speech recognition?

You say "what's the weather tomorrow" into your phone and words pop up on screen. That's speech recognition (ASR) in action. Its job is simple to state and hard to do: turn sound waves in the air into lines of text you can search and process.

How does sound become text?

Slice it up first
Incoming audio is cut into tiny frames of a few dozen milliseconds, each carrying acoustic features like frequency and energy.
An acoustic model guesses the sounds
One model reads those frames and guesses which sounds were likely spoken, producing a string of phonemes.
A language model fixes mistakes
Phones alone aren't enough — too many words sound alike. The language model uses context to decide between lookalikes, then outputs the most likely text.

Where has it changed things?

Voice assistants
Siri and Alexa understand your commands thanks to ASR.
Automatic captions
Subtitles for videos, meetings and live streams are often generated in real time.
Transcription
Interviews and meeting notes no longer need manual playback and typing.
Accessibility
It lets deaf and hard-of-hearing people "see" what others are saying.

What's still hard?

Accents, background noise, overlapping speakers and jargon all cause errors. Still, with more data and compute, recognition accuracy now rivals — or beats — the average human.

Bottom line: speech recognition turns what you say into text a machine can read.