Contextualized Translation of Automatically Segmented Speech
This addresses a practical issue for speech translation systems by improving robustness to real-world audio segmentation, though it is incremental as it builds on existing methods.
The paper tackles the problem of speech-to-text translation quality degradation due to mismatched segmentation between training (sentence-level) and inference (VAD-based), by making the model robust to sub-optimal segmentation. The result is a context-aware solution that outperforms a strong base model by up to 4.25 BLEU points on an English-German test set.
Direct speech-to-text translation (ST) models are usually trained on corpora segmented at sentence level, but at inference time they are commonly fed with audio split by a voice activity detector (VAD). Since VAD segmentation is not syntax-informed, the resulting segments do not necessarily correspond to well-formed sentences uttered by the speaker but, most likely, to fragments of one or more sentences. This segmentation mismatch degrades considerably the quality of ST models' output. So far, researchers have focused on improving audio segmentation towards producing sentence-like splits. In this paper, instead, we address the issue in the model, making it more robust to a different, potentially sub-optimal segmentation. To this aim, we train our models on randomly segmented data and compare two approaches: fine-tuning and adding the previous segment as context. We show that our context-aware solution is more robust to VAD-segmented input, outperforming a strong base model and the fine-tuning on different VAD segmentations of an English-German test set by up to 4.25 BLEU points.