Speech synthesis (TTS)

Language, voice & vision

Speech synthesis, or text-to-speech (TTS), turns text into voice. Recent models produce natural, expressive, cloneable voices in many languages. In a voice agent, it provides the timbre and the embodiment. Choosing the voice is as much a design decision as a technical one.

In practice at Gensai

For the Oceans, Gensai gave the sea a deep, mysterious voice through Gradium's speech synthesis.

How to choose a voice?

Start from the use case: a calm, clear voice for phone reception, a more embodied voice for a cultural experience or a character. Three criteria matter: latency before the first word, naturalness on long sentences and correct pronunciation of proper names and business terms. A French or European solution, such as Gradium, also simplifies compliance questions.

Common mistake

Cloning a voice without a legal framework. Using the voice of a living person requires their explicit consent, and the character must be framed so that the voice never says anything the person or the institution would not stand behind.