REVIEW 2 cited by
DistillW2V2: A Small and Streaming Wav2vec 2.0 Based ASR Model
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Wav2vec 2.0 (W2V2) has shown impressive performance in automatic speech recognition (ASR). However, the large model size and the non-streaming architecture make it hard to be used under low-resource or streaming scenarios. In this work, we propose a two-stage knowledge distillation method to solve these two problems: the first step is to make the big and non-streaming teacher model smaller, and the second step is to make it streaming. Specially, we adopt the MSE loss for the distillation of hidden layers and the modified LF-MMI loss for the distillation of the prediction layer. Experiments are conducted on Gigaspeech, Librispeech, and an in-house dataset. The results show that the distilled student model (DistillW2V2) we finally get is 8x faster and 12x smaller than the original teacher model. For the 480ms latency setup, the DistillW2V2's relative word error rate (WER) degradation varies from 9% to 23.4% on test sets, which reveals a promising way to extend the W2V2's application scope.
Forward citations
Cited by 2 Pith papers
-
Effective and Efficient One-pass Compression of Speech Foundation Models Using Sparsity-aware Self-pinching Gates
A per-layer learnable threshold, trained jointly with CTC fine-tuning, prunes 60-65% of parameters from wav2vec2-base and HuBERT-large without a statistically significant WER increase on test-clean.
-
Unfolding A Few Structures for The Many: Memory-Efficient Compression of Conformer and Speech Foundation Models
A small trained 'seed' speech encoder can be unfolded to several logical depths by repeating shared layers, matching independently trained models with up to 35% parameter reduction.
Discussion (0). Continue with ORCID to comment.