voice analysis
the voice speaks too. without saying a word.
A consultant can say “I’m fine” in a voice that says otherwise. The transcript captures the words; how they were said — the accelerating rhythm, the lengthening pause, the falling tone — is lost unless someone listens for it. nexmin’s voice analysis listens to that layer and returns it as signals with their exact minute.
what is measured, exactly.
Prosody and expressive variability. Speech rate in words per minute. Pauses and response latencies. The share of speech between consultant and therapist. And vocal tension normalised to an index from 0 to 100. These are computational estimates for you to review — not precision instrument readings, and we say so plainly.
what it gives a therapist.
A flattened affect the text does not reveal. Speech that speeds up around a certain topic. The difference between a silence that is doing work and one that is avoiding something. And incongruences — when the voice says the opposite of the words — marked with their minute so you can jump to that point and listen yourself. The machine gives you the signal; the clinical reading is yours.
what nexmin does not do with your voice. deliberately.
nexmin does not recognise or attribute emotions. It does not say that a consultant “is sad”: it documents that the voice falls and latency rises at minute 41, and leaves the meaning to the person with the relationship and context. That line follows the EU AI Act (2024/1689) — detecting perceptible expressions is not inferring inner states — and we treat it as a design principle. No permanent voice identifier is kept either: speaker assignment is proposed for each session and you have the final say.
alongside the rest of the analysis.
The voice does not stand alone: its signals feed into process reading by Pensa and sit alongside process variables and the analysis configured for your practice. When voice and content contradict each other, that contradiction is often the most clinically interesting part of the session — and it stays marked where it happened.
frequently asked questions.
Is this emotion recognition?
No, deliberately. nexmin records acoustic facts — rhythm, pauses and tension — and flags possible incongruences. You decide what emotions may sit behind them. The EU AI Act distinguishes detecting expressions from inferring inner states, and nexmin stays on the right side of that line.
How reliable are these measures?
They are computational estimates meant to guide your listening, not a clinical measurement instrument. Each signal includes its minute so you can check it against the real audio.
Is my voice, or my consultants’ voices, stored as an identifier?
No. There is no permanent voice identifier; speaker assignment is proposed session by session and the therapist confirms it.
listen to what your next session says without words.
14 days free, no card