Short answer
Text-to-speech (TTS) converts written text into spoken audio using generative speech models for accessibility, assistants, and media.
Why it matters
Good TTS unlocks inclusive products; voice cloning extensions require consent and fraud awareness.
Diagram
Article
Use
Great for accessibility and learning.
Examples
Course narration
An instructor pastes lesson text, selects a calm voice, exports MP3, then manually fixes chapter titles misread as acronyms using a pronunciation lexicon.
IVR prompts
A support system generates dynamic balance-readout prompts with SSML for currency amounts, keeping latency low via streaming synthesis.
FAQ
What is SSML?
Speech Synthesis Markup Language is a markup format some TTS engines support to control pauses, say-as interpretation, pitch, and other pronunciation details beyond plain text.
Why does TTS mispronounce names?
Names and brand terms are long-tail linguistic items. Custom lexicons, phonetic spellings, or SSML phonemes often fix recurring errors.
Is streaming TTS different from batch?
Streaming synthesizes and delivers audio chunks with lower time-to-first-byte for interactive apps. Batch generation optimizes for full-file quality/throughput in offline workflows.
Sources
-
Google Cloud Text-to-Speech documentation
GoogleOfficial documentation for a major cloud TTS API.
-
OpenAI platform docs
OpenAIOfficial platform documentation covering speech-related APIs as offered.
-
W3C SSML specification
W3CStandards documentation for speech synthesis markup concepts.