Table 7.
Word error rates (WER) for LittleBeats™ and smartphone audio with and without a language model (Study 5).
Models | LittleBeats™ WER | Smartphone WER |
---|---|---|
Greedy decoding | 5.75% | 3.58% |
Beam search | 5.80% | 3.63% |
Beam search + language model | 4.16% | 2.73% |