Erroneous result for LS test-other
#2
by sanchit-gandhi - opened
Current result is 1.3% WER on LibriSpeech test-other. According to the paper, Whisper base achieves 12.4% WER on test-other (c.f. https://cdn.openai.com/papers/whisper.pdf). This erroneous result is placing Whisper at the top of the HF Speech Bench! (c.f. https://huggingface.co/spaces/huggingface/hf-speech-bench)
sanchit-gandhi changed pull request status to closed