Pin transformers==4.57.6 (CohereAsr/Whisper variants require pre-5.0 native classes)

Add OpenAI gpt-4o-transcribe benchmark on CoRal v3 test

Add ElevenLabs scribe_v2 benchmark on CoRal v3 test

Embed CER ranking charts above the comparison tables

Add benchmark chart: bench-readaloud.png

Add benchmark chart: bench-conversation.png

Add capacit-ai/saga to CoRal v3 CER comparison tables

Add inference speed (RTFx ~425 on RTX 3090, 60min audio in ~8.5s)

Remove obsolete v5.2 comparison

Add CER comparison vs CoRal/roest Danish ASR models on CoRal v3

Remove upstream model card references; describe architecture generically

Relicense to CC BY-NC 4.0; commercial use requires separate license (mads@syv.ai)

Expand README with usage examples (single, long-form, batch, beam, punctuation, transcribe wrapper)

Add beam=5 results: avg WER 15.16% raw / 13.91% strict

Add model card with full-test eval and training details

Upload hviske-v5.3: LLRD mult=0.75, 5 epochs — full-test avg WER 15.56% (read_aloud 10.26%, conv 21.30%)

initial commit