Developing Speech Processing Pipelines for Police Accountability

TL;DR

This paper explores using large pre-trained speech models to automate the review of police body camera footage, focusing on improving accuracy in officer speech recognition and detection amidst noisy multi-speaker environments.

Contribution

It introduces a pipeline for fine-tuning speech models for police footage, demonstrating significant improvements in officer speech recognition accuracy and providing practical guidance for domain adaptation.

Findings

Fine-tuning improves officer speech ASR WER to 12-13%

Officer speech recognition outperforms community speech recognition

Domain-specific tasks like speaker diarization remain challenging

Abstract

Police body-worn cameras have the potential to improve accountability and transparency in policing. Yet in practice, they result in millions of hours of footage that is never reviewed. We investigate the potential of large pre-trained speech models for facilitating reviews, focusing on ASR and officer speech detection in footage from traffic stops. Our proposed pipeline includes training data alignment and filtering, fine-tuning with resource constraints, and combining officer speech detection with ASR for a fully automated approach. We find that (1) fine-tuning strongly improves ASR performance on officer speech (WER=12-13%), (2) ASR on officer speech is much more accurate than on community member speech (WER=43.55-49.07%), (3) domain-specific tasks like officer speech detection and diarization remain challenging. Our work offers practical applications for reviewing body camera footage and general guidance for adapting pre-trained speech models to noisy multi-speaker domains.