davidggphy/

librispeech-arpabet-processed

Pre-processed dataset for training ARPAbet phoneme recognition models using CTC loss.

This dataset is derived from LibriSpeech (train-clean-100 split) with the following preprocessing:

Audio: Resampled to 16kHz, normalized using Wav2Vec2 feature extractor Labels: Text transcriptions converted to ARPAbet phoneme sequences using CMU Pronouncing Dictionary Filtering: Samples with out-of-vocabulary words (not in CMU… See the full description on the dataset page: https://huggingface.co/datasets/davidggphy/librispeech-arpabet-processed.