Text-to-speech (TTS)

Text-to-speech (TTS) covers the technologies that turn written text into audible speech. For a newsroom, it is the building block that makes an audio version of every article possible without booking a studio or a voice actor.

Two generations of technology

The first generation, known as concatenative synthesis, stitched together pre-recorded voice fragments. The output was intelligible but mechanical: flat intonation, audible seams between segments, no adaptation to the meaning of the sentence. That is the voice of old GPS units and phone menus, and the reason automated reading had a poor reputation for years.

The current generation relies on neural networks trained on large volumes of natural speech. The model no longer glues fragments together: it generates the audio signal with the whole sentence in view. It places pauses where a human reader would, rises on a question, slows down for an aside. This ability to shape rhythm and intonation is called prosody, and it is what separates a voice you tolerate from a voice you actually listen to.

What it means for a news publisher

In practice, an editorial-grade TTS system has to handle real newspaper copy: acronyms, place names, figures, quotes, bylines. Preparing the text properly matters as much as generating the audio.

For a regional daily, the real issue is coverage. Publishing dozens of articles a day makes human narration impossible to generalise; TTS lets you cover the full output, from the local council report to the weekend feature. WeDispatch applies this to news sites with an embedded audio player that sits on every article. The question is no longer whether a machine can read a text aloud. It clearly can. The question is what a newsroom does with that capability: accessibility, new reading habits, and revenue.