Model card: concise training description

README.md

@@ -20,7 +20,7 @@ tags:
 
 **Realtime Amharic speech recognition by Addis AI.** Text appears while people speak, in 0.32-second steps,
 and is final about 30 ms after they stop. Built on NVIDIA Nemotron 3.5 ASR Streaming (0.6B), trained on
-5,000+ hours of Amharic in which every audio-transcript pair was machine-verified.
+5,000+ hours of quality-checked Amharic speech.
 
 | | Addis Scribe Streaming | [Google Chirp 3](https://cloud.google.com/speech-to-text/v2/docs/chirp_3-model) (streaming) |
 |---|---|---|
@@ -105,11 +105,8 @@ benchmark clips).
 
 ## Training
 
-Trained on 5,000+ hours of Amharic speech. Before training, every pair was checked: a second recognizer re-transcribed each clip's packed audio
-and any clip whose text disagreed with it (character error above 0.20) was removed, leaving 2,222,624 verified
-clips. Training ran 48,000 steps, then a
-final pass on the 2,408 human-transcribed hours only. FLEURS, WAXAL test and the evaluation speakers were never
-trained on.
+Trained on 5,000+ hours of quality-checked Amharic speech, then refined on human-transcribed speech.
+FLEURS, WAXAL test and the evaluation speakers were never used in training.
 
 ## Limits