Text-to-speech

How it works

How it works

A neural network turns text into sound in the chosen voice and places stress, intonation, and pauses automatically. If the system reads something wrong, SSML markup helps: it sets speed, pauses, and how to read numbers, dates, and names.

• On-the-fly synthesis – for a live conversation with an AI agent
• Pre-generated phrases – for a menu that rarely changes
• SSML markup – to fix pronunciation and pauses

Use

Where it is used

• IVR menus and announcements that change often
• AI voice agents and outbound reminders
• Reading out customer data: amounts, dates, and order numbers
• Announcements in several languages

Practice

What to check

• How the voice sounds in English on your real phrases, not demo ones
• How it reads numbers, addresses, and your product names
• How long it takes until the first sound: in a live conversation a long pause is noticeable
• Terms of use for the voice and where synthesis runs – in the cloud or on your own servers

In CONTACT.WORK

How it works in CONTACT.WORK

Text-to-speech gives voice to the IVR menu and to the AI voice agent's answers, which the agent builds from the knowledge base and CRM data. If a menu changes often, it is easier to synthesize its phrases than to record them every time. With synthesis, the IVR can say any data from your business systems – order status, balance, appointment date – without pre-recorded phrases.

Synthesis runs in the CONTACT.WORK cloud or on your own servers. The platform itself is free. Text-to-speech is paid separately, or the model runs on your own hardware. The server calculator shows which server you need to run speech synthesis yourself.