Whisper-small Fine-Tuned for Air Traffic Control

This model is Whisper-small fine-tuned with LoRA on the ATCOSIM and UWB-ATCC corpora, with the adapter merged into the base weights. On the ATCO2 corpus, which was not used in training, word error rate fell from 0.567 for the pretrained model to 0.296.

The model inherits the non-commercial CC BY-NC-SA 4.0 license of UWB-ATCC. Training, evaluation, and the demo are documented at https://github.com/jsf3467v/whisper-atc-lora.

Downloads last month
37
Safetensors
Model size
0.2B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for jsf3467v/whisper-small-atc-lora

Adapter
(330)
this model

Datasets used to train jsf3467v/whisper-small-atc-lora