Acoustic Model
An ASR model component that maps acoustic observations to probabilities or scores over phonetic, subword, or token representations.
Speech Processing Context
An acoustic model relates observed speech features to linguistic units or token probabilities. In hybrid HMM systems it commonly provides acoustic likelihoods, while end-to-end CTC, transducer, and attention systems blur the boundary between separate acoustic and sequence models.
Modeling Boundary
An acoustic model and a language model encode different sources of information. The former models evidence in the audio signal; the latter scores linguistic sequence plausibility. Noise, channel variation, and speaker diversity strongly affect acoustic-model generalization.