Tone of voice

Most research on reading others has focused on the face, for practical reasons: a photograph is easier to standardise than a recording. The vocal channel has been less studied, even though it is available in a great many situations where the face is not.

The parameters at play are measurable: pitch, pitch variation, intensity, rate, pauses, timbre quality. They combine into profiles that listeners classify with respectable consistency, including across languages.

Studies comparing channels report that listening alone sometimes gives better results than video for identifying an interlocutor's state. The explanation proposed is that the face draws attention and supplies socially controlled cues, whereas the voice is harder to stage over time.

This result calls for caution: it depends greatly on the materials and the instructions. It is nonetheless enough to discard the received idea that the face is the main channel and the voice a supplement.

A telephone conversation removes the face and the body, leaves the voice intact, and preserves the synchronisation of rhythm. It is a configuration poor in channels but rich in the one that remains, and ordinary experience confirms that one perceives very well there that a person is not in their usual state.

Writing removes the channel entirely. That is no detail: the same sentence, read without prosody, lends itself to opposite readings, and the reader projects onto it the tone they expect. That is the main mechanism of messaging misunderstandings.

Some vocal parameters are common to very different states. A rise in pitch accompanies anger as well as intense joy, an acceleration of rate accompanies enthusiasm as well as anxiety. A listener without the context regularly mistakes the sign.

Added to this are stable differences between people: some speak fast and loud in all circumstances. Comparing someone with themselves informs far better than comparing them with a norm, which presupposes knowing them a little.

Automatic analysis of tone has spread in call centres, where it serves to flag a dissatisfied customer or to assess an adviser. It rests on the same contested hypothesis as facial recognition: that a signal profile corresponds reliably to a state.

The European regulation on artificial intelligence adopted in 2024 prohibits emotion inference systems in the workplace, outside medical or safety reasons. The question of the exact scope of that prohibition, for systems claiming to measure customer satisfaction rather than the employee's state, will in all likelihood keep lawyers busy.