Read the same sentence as a question and then as a warning. The transcript may stay the same while the pitch contour, timing and distribution of acoustic energy change. A transcript-only interface discards those measurements before a downstream model can use them.
Spectra-2 exposes a transcript together with acoustic emotion and speaking-style scores. Treat the scores as model outputs and inspect their behavior on representative recordings. A high score for a vocal style does not establish a speaker’s intent or internal state.
In a separate research experiment we used a fruit-fly connectome as a reservoir for an acoustic classification task. That experiment asks which circuit properties help a readout discriminate signals. The original study describes the control reservoirs and the limits of drawing biological conclusions from their performance.
Sources and experiments
Continue exploring
Hear the difference on your own audio.
Try an English recording without an account, inspect the transcript and vocal-expression annotations, and use the quickstart to bring the result into your application.