Fix usage example date

demo/README.md

@@ -1,13 +1,14 @@
 # Demo audio
 
-Same sentences, same reference voice, two models:
+Same sentences, same reference voice, same call, two models:
 
 - `NN_stock.wav` — stock Chatterbox Multilingual v3, given the identical
-  Amharic text with `language_id="en"` (the base model has no Amharic mode
-  and zero Fidel tokenizer coverage; this is what "no support" sounds like).
-- `NN_finetuned.wav` — v3 + this adapter, `language_id="am"`,
-  `temperature=0.6`, `cfg_weight=0.5`.
+  normalized text, no language tag (the base has no Amharic mode and no
+  Fidel tokenizer coverage; this is what "no support" sounds like).
+- `NN_finetuned.wav` — v3 + this adapter through the released loader, no
+  language tag, `temperature=0.6`, `cfg_weight=0.5`.
 
-Texts are in `sentences.txt` (tab-separated: index, text). The reference
-voice is a consenting in-house speaker from the `ih_` corpus. All clips carry
-the PerTh watermark.
+Texts are in `sentences.txt` (tab-separated: index, text). `reference.wav`
+is the voice both models cloned (one of us). All generated clips carry the
+PerTh watermark; `before_after.mp4` is all eight pairs in one video for
+GitHub.