A robot voice, and why we say so
What the AI Audio Preview voice is, how it is made, why it sounds like a robot and why that is the honest starting point for sound from words.
If you press Generate on a voice-over in AI Audio today, you will hear a robot. We would rather you heard that first from us.
How the voice is made#
The Preview voice is built in your browser in plain code, with no model and no recording. A pulse like the one in a human throat is passed through filters tuned to vowel sounds, with noise for the hissing consonants. The pitch comes from the voice you choose, the contour from the emotion, and the pace from the speed. The result is intelligible and mechanical.
What else the Preview makes#
Music beds from chords, a bass line, an arpeggio and drums in the mood you choose, with a loop that folds its tail into its start. Sound effects shaped by the word in your description, such as rain, wind, whoosh, beep, boom, click, laser or chime. Dialogue, where each turn gets its place. All real sound files, all simulated.
Why ship it#
Because the product around it is real, and it needs something to run on. The composer, the credits shown before you press, the takes with waveforms, the history of runs, variations, the library, uploads, downloads and saving to Geenius OS Media with provenance all work with the Preview. When a real voice model is switched on, it plugs into the same place and nothing else changes.
What is connected#
Gemini speech, ElevenLabs voices and an optional OpenAI voice are connected in code for single voices. They work only where their keys are set on the server or in your workspace's Geenius OS secrets, and whether any of them is on for the hosted app is something we say we have to confirm. Google's Lyria for music and ElevenLabs sound effects are planned and not connected, and two real voices in one dialogue are planned too.
How to use the Preview#
Use it as a scratch track, to check timing and length and to try a script in three languages. Replace it before anything goes out. Do not put it in front of a customer as a finished voice.
The point#
A product that tells you what its voice is made of is a product you can trust with the next thing it says.
See what makes the sound.