Incremental Blockwise Beam Search for Simultaneous Speech Translation with Controllable Quality-Latency Tradeoff
Abstract
Blockwise self-attentional encoder models have recently emerged as one promising end-to-end approach to simultaneous speech translation. These models employ a blockwise beam search with hypothesis reliability scoring to determine when to wait for more input speech before translating further.
However, this method maintains multiple hypotheses until the entire speech input is consumed - this scheme cannot directly show a single incremental translation to users. Further, this method lacks mechanisms for controlling the quality vs. latency tradeoff.
We propose a modified incremental blockwise beam search incorporating local agreement or hold-n policies for quality-latency control. We apply our framework to models with limited and full-context encoders, with the latter demonstrating that offline models can be effectively converted to online models.
Experimental results on MuST-C show 0.6-3.6 BLEU improvement without changing latency or 0.8-1.4 s latency improvement without changing quality.