Table 2 Main differences between pipeline ASR and end-to-end ASR in terms of architecture, decoding strategy, input and output data
From: Performance vs. hardware requirements in state-of-the-art automatic speech recognition
| Â | Pipeline ASR | End-to-end ASR |
|---|---|---|
System architecture | multi-component: | single component: |
| Â | acoustic model (usually neural network) | neural network for both acoustic and language modeling |
| Â | language model (usually probabilistic) | + optional more complex language model |
| Â | phonetic dictionary | Â |
Decoding strategy | Weighted Finite State Transducer (WFST) decoder | Connectionist Temporal Classification (CTC) decoder |
| Â | Â | Sequence-to-sequence attention decoder |
Input | always hand-crafted features | raw waveform or sometimes hand-crafted features |
| Â | + optional learned features | Â |
Output | context dependent phones and then words | characters or word-parts |