SPGISpeech: 5,000 hours of transcribed financial audio for fully formatted end-to-end speech recognition

TL;DR

This work reframes speech-to-text as end-to-end transcription with full orthography, motivated by the value of formatting cues in audio. It introduces SPGISpeech, a 5,000-hour earnings-call corpus with fully formatted transcripts, and demonstrates baseline Conformer models achieving competitive CER on orthographic output. The authors provide extensive corpus construction and privacy safeguards, and release the dataset for non-commercial research to advance end-to-end formatted STT. Overall, the study shows that fully formatted orthographic STT is feasible with current architectures and emphasizes the dataset’s potential to improve downstream NLP and domain-specific transcription tasks.

Abstract

In the English speech-to-text (STT) machine learning task, acoustic models are conventionally trained on uncased Latin characters, and any necessary orthography (such as capitalization, punctuation, and denormalization of non-standard words) is imputed by separate post-processing models. This adds complexity and limits performance, as many formatting tasks benefit from semantic information present in the acoustic signal but absent in transcription. Here we propose a new STT task: end-to-end neural transcription with fully formatted text for target labels. We present baseline Conformer-based models trained on a corpus of 5,000 hours of professionally transcribed earnings calls, achieving a CER of 1.7. As a contribution to the STT research community, we release the corpus free for non-commercial use at https://datasets.kensho.com/datasets/scribe.

Figures (1)

• Figure 1: Distribution of Speakers by Global Region. Speaker distribution is estimated in a random sample according to reported region of corporate domicile.