Strongly Incremental Repair Detection
This work addresses incremental speech repair detection for real-time applications like transcription, but it is incremental as it builds on existing methods with specific improvements.
The authors tackled the problem of detecting speech repairs and edit terms in transcripts with minimal latency, presenting STIR, a system that achieves utterance-final accuracy comparable to state-of-the-art incremental methods while offering better incremental accuracy, faster detection times, and lower computational overhead.
We present STIR (STrongly Incremental Repair detection), a system that detects speech repairs and edit terms on transcripts incrementally with minimal latency. STIR uses information-theoretic measures from n-gram models as its principal decision features in a pipeline of classifiers detecting the different stages of repairs. Results on the Switchboard disfluency tagged corpus show utterance-final accuracy on a par with state-of-the-art incremental repair detection methods, but with better incremental accuracy, faster time-to-detection and less computational overhead. We evaluate its performance using incremental metrics and propose new repair processing evaluation standards.