The listening room
Read it. Then hear it.
Selected full-clip comparisons across speaker and text similarity. Listen to A and B, then use the highlighted transcript differences to inspect what changed.
These eight cases were deliberately selected to illustrate different regions of the scatter. They are not a prevalence sample. Scores use original speaker vectors and full ASR transcripts; the audio is a compressed listening preview.
Loading selected comparisons…
Full-clip MP3 previews, 80 kbps mono, with no trimming or gain normalization. A clip may contain more than one speaker. Displayed speaker cosine uses the two explicitly identified centroids, not a calibrated identity probability.