Whisper-Large-v3-CV11 — 11 languages fine-tune, standard architecture (scientific control)
A plain openai/whisper-large-v3 (unmodified architecture) fine-tuned on the 11 languages of the CC0
Whispered corpus. This is the
matched scientific control for burakaydinofficial/whisper-large-v3-mla-cv11 —
trained identically, minus the MHA→MLA conversion — published so the MLA conversion cost is
independently reproducible. No custom code: loads directly in transformers, and — being a plain unmodified Whisper — is convertible for faster-whisper / CTranslate2 / whisper.cpp via their standard converters.
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
model = AutoModelForSpeechSeq2Seq.from_pretrained("burakaydinofficial/whisper-large-v3-cv11") # no trust_remote_code
processor = AutoProcessor.from_pretrained("burakaydinofficial/whisper-large-v3-cv11")
Reproduce the conversion cost
Evaluate this control and whisper-large-v3-mla-cv11 on CommonVoice-17 (scripts/validate.py in the
code repo); the per-language difference is the
conversion cost reported on the MLA card and in docs/results/.
Results (CommonVoice-17 test, greedy, Whisper normalization + Arabic folding; CER for th/zh/ja)
| Lang | this control |
|---|---|
| en | 6.4 WER |
| de | 5.2 WER |
| es | 4.2 WER |
| fr | 9.5 WER |
| ru | 4.3 WER |
| tr | 8.5 WER |
| cy | 35.1 WER |
| ar | 16.0 WER |
| ka | 72.1 WER |
| th | 4.4 CER |
Encoder frozen during fine-tuning; 15,000 steps, warmup+cosine, fp16. Read-speech domain (CommonVoice + FLEURS-validated). "Compression cost" does not apply to this unconverted control.