Decoder. plain-English AI glossary

Text-to-Speech

● Core

Converting written text into natural-sounding audio—read documents aloud, voiceovers, voice assistants.

Think of it like

An actor reading a script into a microphone, but instantly and in any voice.

Example

Google TTS converts "Hello, how are you?" to audio in seconds; can choose voice, speed, gender.

How it actually works

Two-stage: text-to-mel (attention-based model like Tacotron learns to align text to audio features) then mel-to-audio (vocoder like WaveGlow, HiFi-GAN reconstructs waveform). Prosody (pitch, speed, emotion) controllable via attention mask or explicit tokens. Multi-speaker models learn different voices.

For product teams

Accessibility feature; content monetization (audiobooks from text); voice interface.

For engineers

Encoder processes text; decoder attends and generates Mel spectrogram; vocoder converts to waveform.

Inverse of text-to-speech.
Variant: voice cloning.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome