Beam Search
Definition
A decoding strategy that explores multiple candidate transcriptions simultaneously to find the most likely output.
Beam search is a heuristic search algorithm used during the decoding phase of speech recognition. Instead of greedily selecting the single most probable token at each step, beam search maintains a fixed number (the beam width) of the most promising partial hypotheses and extends them in parallel.
A wider beam generally produces more accurate results at the cost of increased computation. Beam search can be combined with a language model to rescore hypotheses, improving fluency and accuracy. Most production ASR systems use beam widths between 4 and 10.
Frequently Asked Questions
How does beam search differ from greedy decoding?
Greedy decoding picks the single best token at each step, while beam search tracks multiple candidates simultaneously, often producing more accurate final transcriptions.