Toward Zero Oracle Word Error Rate on the Switchboard Benchmark

TL;DR

This paper improves the evaluation of speech recognition systems on the Switchboard benchmark by correcting references and scoring methods, achieving near-zero oracle WER and surpassing human performance with research systems.

Contribution

It introduces a refined evaluation scheme, alternative metrics, and oracle WER computation methods that significantly lower error rates and better discriminate between human and machine performance.

Findings

Commercial systems can score below 5% WER with corrected references.

Research systems surpass commercial human speech recognition accuracy.

Oracle WER can be reduced to 0.18% using dense lattices and alternative data structures.

Abstract

The "Switchboard benchmark" is a very well-known test set in automatic speech recognition (ASR) research, establishing record-setting performance for systems that claim human-level transcription accuracy. This work highlights lesser-known practical considerations of this evaluation, demonstrating major improvements in word error rate (WER) by correcting the reference transcriptions and deviating from the official scoring methodology. In this more detailed and reproducible scheme, even commercial ASR systems can score below 5% WER and the established record for a research system is lowered to 2.3%. An alternative metric of transcript precision is proposed, which does not penalize deletions and appears to be more discriminating for human vs. machine performance. While commercial ASR systems are still below this threshold, a research system is shown to clearly surpass the accuracy of commercial human speech recognition. This work also explores using standardized scoring tools to compute oracle WER by selecting the best among a list of alternatives. A phrase alternatives representation is compared to utterance-level N-best lists and word-level data structures; using dense lattices and adding out-of-vocabulary words, this achieves an oracle WER of 0.18%.