Logo Lanfrica

anashas/Automatic-Speech-Recognition-in-WOLOF

Domain:

natural language processing

Record type:

project
Creator:
ana
Host:
### Automatic-Speech-Recognition-in-WOLOF - This notebook is part of my participation in this Zindi competition about Automatic Speech Recognition in WOLOF. - My Zindi Profile - The main goal is to build an ASR model for the WOLOF language - Fine-tuned the ```facebook/wav2vec2-large-xlsr-53``` model from ```Hugging Face``` Model Hub - Used ```datasets``` from the ```transformers``` library to preprocess the transcriptions and ```torchaudio``` to resample the audio files to 16 kHz - Model hyperparameters: ```attention_dropout=0.05, hidden_dropout=0.05, feat_proj_dropout=0.0, mask_time_prob=0.05, layerdrop=0.1,learning_rate=1e-4``` - The ```Automatic_Speech_Recognition_in_WOLOF.ipynb``` notebook contains the step for training and submitting your model