Real vs synthetic: listen side by side

Real podcast and real Aloe recordings (held-out: no model trained on them), next to synthetic persona conversations voiced by the final delivery model (v4 + Aloe, w-114, settings FY4). Stereo: one speaker per side. Synthetic audio stops at the end of the script.

Real podcast conversations

Held-out podcast conversations, the first 10 minutes of the original recording, with the transcript.

Real Aloe conversations

Held-out Aloe sessions (two strangers on a call), first 8 minutes, both speakers' separated tracks.

Synthetic conversations voiced by the final model

Two per activity from the 4,584 voiced so far, preferring ones whose script was delivered completely (most were not: see the status note). Each shows the script both copies of the model read and what each side actually said, with times.