StableQuant adaptively clips outliers per layer before choosing quantization scales, keeping WER within 0.3% at 8-bit precision on HuBERT and wav2vec2.0.
wav2vec: Unsupervised pre-training for speech recognition,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
StableQuant: Layer Adaptive Post-Training Quantization for Speech Foundation Models
StableQuant adaptively clips outliers per layer before choosing quantization scales, keeping WER within 0.3% at 8-bit precision on HuBERT and wav2vec2.0.