Lesson 55 · 2 min
AI that hears: speech to text
Speech-to-text AI turns recorded or live talk into written words, which makes notes, captions and search possible.
Speech-to-text models were trained on huge amounts of audio paired with transcripts. They learn which sounds match which words, and use the surrounding words to pick between similar ones.
Today they power phone dictation, video captions and meeting notes. Many handle dozens of languages, including Vietnamese, and cope with background noise far better than older systems.
They still slip. Names, numbers, technical words and strong accents cause most errors. In Vietnamese, a missed tone can turn one word into a very different one.
Treat a transcript as a draft. Before you share meeting notes or quote someone, check the names, figures and anything that sounds odd against the recording.
Do it now
Dictate and check
- 1. Open the voice typing on your phone keyboard or in a notes app.
- 2. Read out a short message that includes a person's name, a phone number and a price.
- 3. Look at which parts came out wrong. Those are the parts to check in every transcript.
Quick check
Keep going
Next →AI that speaks: text to speechWords from this lesson
- Speech-to-text
- AI that turns spoken words into written text, as in meeting transcripts and voice typing.
- NLP
- The field of computing that deals with understanding and producing human languages like English, Hindi or Swahili.