Skip to content
Limited-time offer. Save up to 50% on Speechly Premium. See pricing

Text-to-Speech Glossary: 40 Terms Explained

A text to speech glossary of 40 terms in plain English: neural voices, prosody, SSML, vocoders, OCR, on-device vs cloud and more, with real examples.

Updated 6 min readBy the Speechly team
A grid of forty small white paper tiles with embossed shapes, one magenta tile lifted out of the grid.
Quick answer

This text to speech glossary defines 40 terms in plain English. The core ones: text to speech (TTS) turns written text into spoken audio; a neural voice uses machine learning to sound natural; prosody is the rhythm and melody of speech; and SSML is markup that tells a voice how to say things.

Key takeaways
  • Text to speech has three stages: cleaning up the text, deciding how it should sound, and generating the audio. Most terms describe one of these stages.

  • Neural voices replaced older concatenative and formant voices because they handle rhythm and intonation far more naturally.

  • Prosody, the pitch, timing and stress of speech, is what separates a voice you can listen to for an hour from one that sounds robotic.

  • On-device voices make speech on your phone or computer; cloud voices send text to a server. That affects privacy and whether you need internet.

  • File terms like OCR, reflowable text and DRM explain why some documents read cleanly and others don't.

This text to speech glossary explains 40 terms you'll meet in app settings, product pages and articles about voices, in plain English. Each definition says what the term means and why it matters when you're actually listening.

The terms follow the path text takes: from the basics, through how a voice turns text into sound, to how it sounds, how you control it, the files it reads, and where the speech is made. For the full picture in one read, start with what text to speech is and how it works.

Text to speech glossary: the basics

  1. Text to speech (TTS). Technology that reads written text aloud in a synthetic voice. It powers everything from your phone's read-aloud feature to audiobook-style reading apps.
  2. Speech synthesis. The technical name for generating speech by computer. "Speech synthesizer" and "TTS engine" usually mean the same thing.
  3. TTS engine. The software that converts text into speech. On Android you choose a preferred engine in Settings > Accessibility > Text-to-speech output (Google).
  4. Voice. One particular speaker an engine can produce, with its own sound, gender presentation, accent and language. One engine can offer dozens of voices.
  5. Read aloud. A general name for features that speak the text in front of you, such as Read aloud in Microsoft Edge.
  6. Screen reader. Software that reads and navigates a device's whole interface by voice, including buttons and menus. VoiceOver, TalkBack and Narrator are screen readers built mainly for blind and low-vision users.
  7. Speak Selection / Select to Speak. Features that read only the text you highlight. Apple calls it Speak Selection; Android and Chromebooks call it Select to Speak.
  8. Speech to text (STT). The reverse of TTS: turning spoken words into written text, as in dictation. Also called speech recognition or ASR (automatic speech recognition).

How a voice turns text into sound

  1. Text normalization. The first step, where the engine rewrites text into speakable words. "Dr." becomes "doctor" or "drive", "$5" becomes "five dollars", and "1990" becomes "nineteen ninety".
  2. Phoneme. The smallest unit of sound that changes meaning in a language, such as the difference between "bat" and "pat". Engines plan speech as sequences of phonemes.
  3. Grapheme-to-phoneme conversion (G2P). The step that works out how written letters should sound. It's why an engine knows "tough" and "though" are pronounced differently.
  4. Lexicon. A pronunciation dictionary the engine checks before guessing. Custom lexicons fix brand names, surnames and technical terms.
  5. Homograph. A word spelled the same but said differently depending on meaning, like "read" (now) and "read" (past), or "lead" the metal and "lead" the verb. Homographs are a classic source of TTS mistakes.
  6. Formant synthesis. An early method that generates speech from rules modeling the human vocal tract, with no recorded voice. Easy to understand, but unmistakably robotic.
  7. Concatenative synthesis. A method that stitches together tiny recorded snippets of a real speaker. More human than formant voices, but you often hear the joins.
  8. Parametric synthesis. A statistical method that generates speech from a model of a voice instead of stitching recordings. Smoother than concatenative voices, though often muffled or buzzy.
  9. Neural TTS. Speech generated by deep learning models trained on large amounts of recorded speech. It's what makes modern voices sound natural; how AI voices work explains it step by step.
  10. Acoustic model. The part of a neural system that predicts what the speech should sound like, including timing and pitch, before any audio exists.
  11. Mel spectrogram. A picture of sound over time, scaled to match how human hearing works. Many neural systems predict a mel spectrogram first, then turn it into audio.
  12. Vocoder. The component that turns that predicted representation into an actual audio waveform. DeepMind's WaveNet, introduced in 2016, was a landmark neural approach to generating audio.

How a voice sounds

  1. Prosody. The melody and rhythm of speech: pitch, timing, pauses and stress together. Poor prosody is the main reason a voice sounds flat; more in why some voices sound robotic.
  2. Intonation. How pitch rises and falls across a sentence. It's how you hear the difference between a statement and a question.
  3. Pitch. How high or low a voice sounds. Many devices let you adjust it, though changing it too far makes voices sound unnatural.
  4. Stress. The extra emphasis on certain syllables or words. Misplaced stress can change meaning, as in "REcord" (a noun) and "reCORD" (a verb).
  5. Speaking rate. How fast the voice talks, often shown as words per minute or as a multiplier.
  6. Speed multiplier (1×, 1.5×, 2×). Playback speed relative to the voice's normal pace. 2× is twice as fast as normal. Higher speeds stay clearer with some voices than others.
  7. Locale. A language plus a region, such as English (US), English (UK) or Portuguese (Brazil). It decides the accent, spelling conventions and how dates and numbers are read.
  8. Enhanced and Premium voices. Apple's names for higher-quality versions of its built-in voices, downloaded in Settings > Accessibility > Read & Speak > Voices on iPhone. They take more storage but sound much less robotic.
  9. Expressive voice. A voice that varies emotion and emphasis with the text, rather than reading everything in one even tone. Good for fiction, less essential for textbooks.

Controls and markup

  1. SSML (Speech Synthesis Markup Language). A W3C standard for tagging text with instructions: pause here, say this as a date, spell this out. Our SSML explainer shows examples.
  2. Custom pronunciation. A setting where you teach a voice how to say a word. On iPhone and Mac it's Pronunciations in Read & Speak.
  3. Word highlighting. Highlighting each word as it's spoken, also called synchronized or karaoke-style highlighting. It helps you follow along and keep your place.
  4. Web Speech API. A browser standard that lets web pages use text to speech. Microsoft says Edge's Read aloud uses it.

Files and formats

  1. OCR (optical character recognition). Software that recognizes text in images, such as a photo of a book page or a scanned PDF, so it can be read aloud. See what OCR is.
  2. Reflowable text. Text that rewraps to fit any screen, as in most eBooks and web pages. It's the easiest kind for TTS to read in the right order.
  3. Fixed layout. Pages with locked positions, like most PDFs, comics and textbooks. Columns, sidebars and footnotes can make a voice read things out of order.
  4. EPUB. The standard open eBook format, maintained by the W3C. DRM-free EPUB files open in many reading and TTS apps.
  5. DRM (digital rights management). Copy protection on purchased eBooks and audiobooks. It usually limits a book to the store's own apps, so third-party TTS apps can't open it.

Where speech is made

  1. On-device vs cloud TTS. On-device voices make speech on your phone or computer; cloud voices send your text to a server and stream audio back. Microsoft, for example, says Edge voices with "Online" in their names use its cloud service. Read what to check before you upload for the privacy side.
  2. Voice cloning. Creating a synthetic copy of a specific person's voice from recordings. Apple's Personal Voice is an accessibility version: you read a series of sentences aloud, and your device builds a voice that sounds like you, processing it on the device.

Putting the terms to work

Most of these terms come down to two practical choices: which voice you listen to, and whether the app can read your files cleanly.

  • Choosing a voice: look for neural or Premium voices with good prosody in your locale, and test them at your usual speed multiplier.
  • Choosing an app: check for OCR if you read printed pages, sensible handling of fixed-layout PDFs, word highlighting, and whether speech is on-device or in the cloud.

Speechly is a text-to-speech reader for iPhone, iPad, Android, Mac and Windows that applies several of these ideas. It has 25+ natural AI voices in eight languages, word-by-word highlighting, speeds from 0.5× to 2×, and photo reading for printed pages. It opens PDFs, Word documents, DRM-free EPUB eBooks and web pages, and can skip repeating PDF headers and footers on iPhone and iPad. Speechly does not offer voice cloning. Download Speechly free; the free plan includes two files with premium voices.

Frequently asked questions

What does TTS stand for?

TTS stands for text to speech: technology that reads written text aloud in a synthetic voice. It's the opposite of speech to text, which turns spoken words into written text.

What is the difference between a TTS engine and a TTS voice?

The engine is the software that does the work of turning text into speech. A voice is one particular speaker the engine can produce, with its own sound, accent and language.

What is a neural voice?

A neural voice is a synthetic voice generated by machine-learning models trained on recorded speech. It sounds far more natural than older voices built from recorded snippets or rules.

What is prosody in text to speech?

Prosody is the melody and rhythm of speech: pitch, timing, pauses and stress. Good prosody makes questions rise, lists sound like lists and long sentences easy to follow.

What is the difference between a screen reader and text to speech?

Text to speech reads text aloud. A screen reader uses text to speech to read and navigate a device's whole interface, including buttons and menus, mainly for blind and low-vision users.

Is SSML the same as text to speech?

No. SSML is a markup language that gives a text-to-speech voice instructions, such as where to pause or how to say a date. Developers and content creators use it; everyday listeners rarely need it.

Speechly

Listen with Speechly

Natural AI voices on iPhone, iPad, Android, Mac and Windows, with word highlighting and your place saved on every device. Free to download.

Prices and features of other products were checked on November 18, 2026 and can change. Speechly is not affiliated with the other companies named here.