Not a concept. A proprietary method, already in production.

I — A perceptual diagnosis, not a technical one

Contemporary voice synthesis has solved the problem of phonetics — articulatory accuracy, signal cleanliness, language coverage — but not that of vocal phenomenology. Between a voice that pronounces a sentence correctly and a voice that inhabits it there is a perceptual threshold, subtle and sharp, recognisable within a few hundredths of a second by any listener. The most recent neural architectures shorten this distance but do not erase it: the flaw is not computational in nature. It is artistic.

Walter Murch and Michel Chion, in different traditions, insisted on the same point: spoken sound is not transported information, it is gesture. The synthetic-voice industry keeps treating it as information. ANIMA starts here — from the recognition that the emptiness the audience perceives is not a training artefact, but the absence of a direction.

II — A formalisable grammar of voice direction

ANIMA's original contribution is not a model, not an engine, not an algorithm. It is a grammar of voice direction that lives above any generative system and redesigns its use. Four categories define its syntax: direction, the act of fixing intention before generating; take, generation treated as raw material to be chosen and not as a result to be accepted; refine, the subsequent work on prosody, weight, ending, breath; retake, the refusal of compromise, the deliberate regeneration.

These categories are not borrowed by analogy with cinematic post-production: they are formally transferred from the mix room, where they have operated identically for decades, into the digital space, where they had never been codified as method. The grammar so described is legible, transferable, teachable. It does not depend on the engine. It survives a change of model.

III — A research substrate, not an intuition

This grammar is not a theoretical position. It is the result of an experimental protocol conducted on a continuous corpus of synthetic vocal emissions, listened to, measured, compared. The method proceeds in cycles: hypothesis on the encoding of a parameter, generation, acoustic analysis, perceptual verification, redefinition.

From this work comes an internal prosodic dictionary — a precision instrument, not a user manual — that maps the model's behaviour with respect to punctuation, descriptors, scene markers and execution directives. The object's status is scientific: hypothesis, replicability, falsifiability. The dictionary is not exposed, because it is the proprietary heart of the method, but its existence defines the difference between a voice-direction system and a collection of tips.

IV — A new category, not a competitor

ANIMA does not present itself as a better voice-synthesis system. It presents itself as an object of a different category: a shaping desk that sits between the generative engine and the usable output. The distinction is structural. A TTS produces voice. A desk shapes it.

The consequence is that synthetic voice leaves the dominant industrial regime — more voices, more languages, more speed — and enters an artisanal-formalisable regime, compatible with professional production as it has been practised for decades in the mix room. It is not a denial of artificial intelligence. It is the recognition that, to cross the perceptual threshold, artificial intelligence needs a craft to receive it, listen to it, correct it. ANIMA is that craft made system.