Aller au contenu
ModelPrésentéExterne
Télécharger

Whisper Base 24lang

A plain openai/whisper-base (unmodified architecture) fine-tuned on the 24 languages of the CC0 Whispered corpus. This is the matched scientific control for burakaydinofficial/whisper-base-mla-24lang — trained...

Sourcehuggingface.co/burakaydinofficial/whisper-base-24lang/tree/6f73aa0251fc5d398432f7ecfa11c90cdc3d7333
TypeText modelModèle de baseopenai/whisper-baseVersionmodelCClasse

Source du modèle

Type
Text model
Taille du fichier
138 MB
Modèle de base
openai/whisper-base
Version
vmodel
Mis à jour
27 sept. 2026
Téléchargements
7
Source
Hugging Face

Description de la source

Whisper-Base-24Lang — 24 languages fine-tune, standard architecture (scientific control)

A plain openai/whisper-base (unmodified architecture) fine-tuned on the 24 languages of the CC0 Whispered corpus. This is the matched scientific control for burakaydinofficial/whisper-base-mla-24lang — 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-base-24lang")   # no trust_remote_code
processor = AutoProcessor.from_pretrained("burakaydinofficial/whisper-base-24lang")

Reproduce the conversion cost

Evaluate this control and whisper-base-mla-24lang 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)

Reconstructed numbers. This control's per-language WER/CER in the table are reconstructed as (the MLA model's absolute score − the measured paired conversion cost), not from a separate evaluation of this checkpoint; an independent re-eval may differ by a few tenths. The paired conversion cost is the directly measured quantity.

Langthis control
en20.5 WER
de29.7 WER
es18.8 WER
fr32.9 WER
it30.5 WER
pt28.4 WER
ru28.0 WER
nl26.9 WER
pl34.4 WER
id38.6 WER
tr38.8 WER
hi36.3 WER
ms35.0 WER
sv-SE43.2 WER
th23.9 CER
zh-CN25.2 CER
cs53.2 WER
vi43.1 WER
fi51.0 WER
el51.6 WER
da56.0 WER
ja34.3 CER
nn-NO67.3 WER
ko62.2 WER

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.

Sources

1 source
Hugging FacePrincipalVersion 6f73aa0251fc5d398432f7ecfa11c90cdc3d7333
huggingface.co/burakaydinofficial/whisper-base-24lang/tree/6f73aa0251fc5d398432f7ecfa11c90cdc3d7333

Vérifié 30 août

Artefacts du modèle

1 artefact

model.safetensors

safetensors · 138 MB · SHA-256 25e0702a3c10…6e7c · Hugging Face

Download file

Extraits de sources

2 extraits
Readme30 aoûthuggingface.co/burakaydinofficial/whisper-base-24lang

--- license: apache-2.0 base_model: openai/whisper-base datasets: - burakaydinofficial/Whispered language: [en, de, es, fr, it, pt, ru, nl, pl, id, tr, hi, ms, sv, th, zh, cs, vi, fi, el, da, ja, "no", ko] pipeline_tag: automatic-speech-recognition library_name: transformers tags: [whisper, multilingual, speech-recognition, fine-tuned, control] --- # Whisper-Base-24Lang — 24 languages fine-tune, standard architecture (scientific control) A **plain `openai/whisper-base`** (unmodified architecture) fine-tuned on the 24 languages of the CC0 [Whispered](https://huggingface.co/datasets/burakaydinofficial/Whispered) corpus. This is the **matched scientific control** for [`burakaydinofficial/whisper-base-mla-24lang`](https://huggingface.co/burakaydinofficial/whisper-base-mla-24lang) — 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. ```python from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor model = AutoModelForSpeechSeq2Seq.from_pretrained("burakaydinofficial/whisper-base-24lang") # no trust_remote_code processor = AutoProcessor.from_pretrained("burakaydinofficial/whisper-base-24lang") ``` ## Reproduce the conversion cost Evaluate this control **and** `whisper-base-mla-24lang` on CommonVoice-17 (`scripts/validate.py` in the [code repo](https://github.com/burakaydinofficial/Whispered)); 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) > **Reconstructed numbers.** This control's per-language WER/CER in the table are reconstructed as (the MLA model's absolute score − the measured paired conversion cost), not from a separate evaluation of this checkpoint; an independent re-eval may differ by a few tenths. The paired conversion cost is the directly measured quantity. | Lang | this control | |---|---| | en | 20.5 WER | | de | 29.7 WER | | es | 18.8 WER | | fr | 32.9 WER | | it | 30.5 WER | | pt | 28.4 WER | | ru | 28.0 WER | | nl | 26.9 WER | | pl | 34.4 WER | | id | 38.6 WER | | tr | 38.8 WER | | hi | 36.3 WER |...

Note30 aoûthuggingface.co/burakaydinofficial/whisper-base-24lang

Source context: 7 downloads · 0 likes · Pipeline automatic-speech-recognition · Library transformers · Repo burakaydinofficial/whisper-base-24lang