banglabridge
/

large-v3-turbo-bn

Automatic Speech Recognition

Eval Results (legacy)

Model card Files Files and versions

This model was finetuned on 21.2k training samples from Common Voice 17.0 bengali subset.

Eval Results on Common Voice 17.0 bengali test split:

Baseline model

Word Error Rate: 118.3%
Loss: 1.074

Finetuned model

Steps: 2000
Word Error Rate: 23.7%
Loss: 0.109

Downloads last month: 27

Safetensors

Model size

0.8B params

Tensor type

F32

·

Model tree for banglabridge/large-v3-turbo-bn

Base model

openai/whisper-large-v3

Finetuned

openai/whisper-large-v3-turbo

Finetuned

(512)

this model

Dataset used to train banglabridge/large-v3-turbo-bn

Evaluation results

wer on Common Voice 17.0
self-reported

23.700