SpeechStew

SpeechStew is a benchmark dataset that contains a variety of speech recognition data used to evaluate the effectiveness of different training objectives, such as the O-1 self-training approach.

Papers using SpeechStew (3)

MAESTRO: Matched Speech Text Representations through Modality Matching2022Accelerating RNN-T Training and Inference Using CTC guidance2022O-1: Self-training with Oracle and 1-best Hypothesis2023