How does text to speech work?
The W3C’s Speech Synthesis Markup Language (SSML) describes the steps a synthesiser takes. It works out the structure of the text, normalises it into words to speak, so “$200” becomes “two hundred dollars”, converts the words into phonemes, plans the prosody (pitch, timing, pauses and emphasis) and produces the audio.
Older synthesisers joined recorded fragments or followed rules, while most current ones use neural networks to generate the audio. SSML lets authors guide the result with markup for pronunciation, pauses, emphasis, speaking rate and pitch. In the browser, the Web Speech API’s SpeechSynthesis interface reads text aloud with the voices available on the device.
Example: the SSML specification notes that “1/2” could be read as “half”, “January second”, “February first” or “one of two”, which is why dates, prices and version numbers are worth testing aloud.
Why does text to speech matter for a voice agent?
The voice is the first thing callers judge. In a chained voice agent, TTS speaks every reply, so its pace, clarity and handling of product names shape the whole call. Text written for a screen, with tables, links and long sentences, often sounds heavy when read out.
Synthetic voices should be disclosed. OpenAI’s text-to-speech guide says its usage policies require a clear disclosure to end users that the voice they hear is AI-generated and not a human.
How does Kawl Now handle text to speech?
A Kawl Now agent’s own speech comes straight from OpenAI’s GPT-Live voice model, not a separate text-to-speech engine, in one of two voices: Marin, warm and expressive, or Cedar, calm and grounded.
There’s no SSML, pronunciation dictionary or speaking-rate setting. Live captions show the words the agent spoke, so a test call is the way to check how it says your product’s names.
Callers are told before the call that they’re speaking with AI, and the agent is told to say it’s an AI. Kawl Now isn’t a text-to-speech tool: it doesn’t read your pages aloud or make audio files for you to use.
Common questions
What is the difference between text to speech and speech to text?
They run in opposite directions. Text to speech turns written words into audio, and speech to text, also called automatic speech recognition, turns spoken audio into written words. A chained voice agent uses both: one to hear the caller and one to reply.
Is text to speech the same as voice cloning?
No. Text to speech can use any voice, including stock voices made by the provider. Voice cloning is a use of speech synthesis that copies a particular person’s voice from recordings of them, which raises questions of consent.