What is text-to-speech?
Text-to-speech (TTS) takes written words and turns them into spoken audio. It is the opposite of automatic speech recognition, which turns speech into text. Together they let a computer listen and talk on the phone.
In a contact centre, TTS gives a voice to IVR menus, AI voice assistants and AI receptionists. Whenever a system reads out a balance, a reference number or an estimated restoration time, that’s usually text-to-speech at work.
How text-to-speech works
Modern TTS uses neural networks trained on many hours of recorded human speech. Given a piece of text, the system:
- Normalises the text, deciding how to say things like “$45.20”, “10/11”, “St” (street or saint?) and “ETR”.
- Works out pronunciation, turning words into the sounds that make them up.
- Plans the delivery, choosing rhythm, pauses, emphasis and intonation (called prosody), so a question sounds like a question.
- Generates the audio in a chosen voice.
This all needs to happen very quickly. On a live call, a noticeable delay between the caller finishing and the AI replying feels unnatural, so response speed (latency) is as important as voice quality.
Why it matters to the customer experience
The voice is the part of an AI system the caller actually experiences. Even if the AI understands perfectly and gives the right answer, a poor voice can undermine trust. Things callers notice include:
- Naturalness. Flat, robotic delivery makes people less patient.
- Clarity on the phone. Phone audio is compressed, so voices need to remain clear on a poor mobile line.
- Pronunciation. Mispronouncing a street, suburb or the customer’s own name is jarring. In New Zealand and Australia, te reo Māori and Aboriginal place names need particular care.
- Numbers. Reading account numbers, dates and phone numbers at the right pace, grouped the way people expect.
- Tone. A cheerful voice reading bad news during an outage feels wrong. Calm and clear usually works best.
TTS vs recorded prompts
Traditional phone systems used pre-recorded prompts from a voice artist. They sound natural but are static, so every wording change means a new recording session. TTS can say anything at any time, which is what makes live information and natural conversation possible. Some organisations still use recorded audio for fixed greetings and TTS for everything dynamic.
Good practice
- Choose a voice that suits your brand and your callers, and use it consistently.
- Build a pronunciation list of local place names, product names and abbreviations, and test each one.
- Write for the ear, not the eye. Short sentences, one idea at a time and key information repeated where it helps.
- Be honest. A natural voice is not a reason to hide that the caller is talking to an automated assistant. Introduce it as one.
- Listen to real calls after launch to catch phrases that sound wrong in practice.