Text to speech (TTS): meaning and example

TTS powers voiceovers, audiobooks, screen readers and voice assistants. Older systems stitched together recorded fragments or used rule-based synthesis and sounded robotic. Today's neural TTS models learn rhythm, stress and intonation from many hours of recorded speech, so they can sound natural and expressive.

How it works

A TTS system works in stages. First, text normalisation turns the written input into speakable words: '£5' becomes 'five pounds' and 'Dr' becomes 'doctor' or 'drive' depending on context. Next, the text is converted into phonemes, the basic sound units of a language, using a pronunciation dictionary and learned rules. An acoustic model then predicts how those sounds should be spoken, including pitch, timing and emphasis, often producing a spectrogram, a picture of sound frequencies over time. Finally, a vocoder turns that spectrogram into an audio waveform you can play. Many modern systems merge these steps into one neural network trained end to end on recorded speech paired with transcripts. Output is commonly delivered at sample rates such as 22 to 48 kHz. Users can usually choose a voice, speed and style, and some tools accept SSML, a markup standard for adding pauses, emphasis and pronunciation hints. Streaming TTS can start speaking within a fraction of a second, which makes real-time voice assistants possible.

Why it matters

TTS lets one person produce narration without a studio, a voice actor or retakes. Creators can voice explainer videos, businesses can localise training content into many languages, and publishers can offer audio versions of articles. It is also a core accessibility tool for people with visual impairments or reading difficulties. The main skill shift is from recording to editing: you fix pronunciation, pacing and emphasis in text rather than re-recording.

Where you will see it

• Narrating YouTube explainers, shorts and course videos.
• Creating audio versions of blog posts and newsletters.
• Reading screen content aloud for blind and low-vision users.
• Giving phone systems and voice assistants a spoken response.

Limits and common misconceptions

• Names, acronyms and specialist terms are often mispronounced unless you add phonetic spellings or pronunciation rules.
• Long narration can sound flat or oddly stressed; emotional delivery still needs careful direction and editing.
• Some platforms require labelling of synthetic voices, and using a voice that imitates a real person needs their consent.

How it differs from related terms

• Text to speech (TTS) vs Speech to text Speech to text does the reverse job: it listens to audio and writes down the words.
• Text to speech (TTS) vs Voice cloning Voice cloning creates a voice that sounds like one specific person; TTS can use any voice, stock or cloned.

Tools that use it

• ElevenLabs
• Murf AI

Frequently asked questions

Related terms

• Temperature A generation setting that controls how random an AI model's word choices are: low values give predictable output, high values give varied output.
• Text to image Generative AI that turns a written prompt into a new image by learning how words relate to visual features.
• Text to video Generative AI that creates moving video clips from a written prompt, and often from a starting image as well.
• Token The basic unit of text a language model reads and writes, usually a whole word, part of a word, a punctuation mark or a space.
• Training data The collection of examples, such as text, images, audio or code, that a machine learning model learns patterns from during training.

All AI terms