arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2502.20583cs.LGcs.AIcs.SDeess.AS
收藏

LiteASR:基于低秩近似的高效自动语音识别

LiteASR: Efficient Automatic Speech Recognition with Low-Rank Approximation

University of Washington(华盛顿大学);Kotoba Technologies, Inc.(Kotoba技术公司)

Keisuke Kamahori, Jungo Kasai, Noriyuki Kojima, Baris Kasikci

AI总结:

针对现代ASR模型编码器计算强度高的部署瓶颈,提出LiteASR低秩压缩方案,通过PCA利用中间激活低秩特性压缩线性变换,将Whisper large-v3编码器压缩超50%且准确性更优,建立了新帕累托前沿。

摘要:

现代自动语音识别(ASR)模型(如OpenAI的Whisper)依赖深度编码器-解码器架构,其编码器因计算强度高而成为高效部署的关键瓶颈。我们提出LiteASR,一种针对ASR编码器的低秩压缩方案,可显著降低推理成本同时保持转录准确性。我们的方法利用中间激活中观察到的强低秩特性:通过在小型校准数据集上应用主成分分析(PCA),用低秩矩阵乘法链近似线性变换,并进一步优化自注意力以在降维空间中工作。评估结果显示,我们的方法可将Whisper large-v3的编码器大小压缩50%以上,匹配Whisper medium的大小且转录准确性更优,从而建立了准确性和效率的新帕累托前沿。LiteASR的代码可在https://github.com/efeslab/LiteASR获取。

英文摘要:

Modern automatic speech recognition (ASR) models, such as OpenAI's Whisper, rely on deep encoder-decoder architectures, and their encoders are a critical bottleneck for efficient deployment due to high computational intensity. We introduce LiteASR, a low-rank compression scheme for ASR encoders that significantly reduces inference costs while maintaining transcription accuracy. Our approach leverages the strong low-rank properties observed in intermediate activations: by applying principal component analysis (PCA) with a small calibration dataset, we approximate linear transformations with a chain of low-rank matrix multiplications, and further optimize self-attention to work in reduced dimensionality. Evaluation results show that our method can compress Whisper large-v3's encoder size by over 50%, matching Whisper medium's size with better transcription accuracy, thereby establishing a new Pareto frontier of accuracy and efficiency. The code of LiteASR is available at https://github.com/efeslab/LiteASR.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑