MedicalAsrDecoderType

Enums · AiDotNet.Enums

Which end-to-end decoder a medical-conversation ASR model uses.

How It Works

Chiu et al., “Speech recognition for medical conversations” (Interspeech 2018, arXiv:1711.07274) is a COMPARISON: “We explored both CTC and LAS systems for building speech recognition models.” Offering both is therefore what implementing that paper means; offering only one reproduces half of it.

Their finding decides the default: “The LAS was more resilient to noisy data and CTC required more data clean up.” Doctor-patient conversation is exactly the noisy, spontaneous condition where that mattered.

Fields

Field Summary
MedicalAsrDecoderType Ctc Connectionist Temporal Classification: a frame-synchronous head over the encoder, with no decoder state.
MedicalAsrDecoderType ListenAttendSpell Listen, Attend and Spell: a bidirectional LSTM encoder (“listen”) feeding a multi-headed attention decoder (“attend and spell”).
int value__