In plain English
Voice AI has three parts: speech to text, text to speech and live voice models. Delay, noise, accents and numbers are the main practical problems. Read key facts back to the caller. Record only with consent, tell people they speak to an AI, and never trust a voice as proof of identity.
- Speech to text
- Turning spoken audio into written words.
- Text to speech
- Turning written words into a spoken voice.
- Voice cloning
- Copying a person's voice from a sample.
Picture it
Step by step, or all in one
Three steps in a chain
- Speech to text, then a language model, then text to speech
- Easy to check and fix each step
- Adds delay at every step
- Most common today
One live voice model