标题： 基于音频-文本的关键词 spotting 的匹配隐编码 显示英文标题

标题： Matching Latent Encoding for Audio-Text based Keyword Spotting

摘要： 使用音频和文本嵌入联合进行关键词检测（KWS）已显示出高质量的结果，但如何语义对齐不同序列长度的多词关键词的两种嵌入仍然是一个未解决的关键挑战。 在本文中，我们提出了一种基于音频-文本的端到端模型架构，用于灵活的关键词检测（KWS），该架构建立在学习到的音频和文本嵌入之上。 我们的架构采用了一种基于新型动态规划算法的动态序列划分（DSP），利用口语内容的单调对齐方式，将音频序列最佳地划分为与基于单词的文本序列相同长度。 我们提出的模型包括一个编码块以获取音频和文本嵌入，一个投影块以将各个嵌入投影到公共潜在空间，以及一个包含新型DSP算法的音频-文本对齐器，该算法对齐音频和文本嵌入以确定口语内容是否与文本相同。 实验结果表明，我们的DSP比其他分区方案更有效，并且所提出的架构在公开数据集上以ROC曲线下的面积（AUC）和等错误率（EER）衡量时，分别优于最先进的结果14.4%和28.9%。 显示英文摘要

摘要： Using audio and text embeddings jointly for Keyword Spotting (KWS) has shown high-quality results, but the key challenge of how to semantically align two embeddings for multi-word keywords of different sequence lengths remains largely unsolved. In this paper, we propose an audio-text-based end-to-end model architecture for flexible keyword spotting (KWS), which builds upon learned audio and text embeddings. Our architecture uses a novel dynamic programming-based algorithm, Dynamic Sequence Partitioning (DSP), to optimally partition the audio sequence into the same length as the word-based text sequence using the monotonic alignment of spoken content. Our proposed model consists of an encoder block to get audio and text embeddings, a projector block to project individual embeddings to a common latent space, and an audio-text aligner containing a novel DSP algorithm, which aligns the audio and text embeddings to determine if the spoken content is the same as the text. Experimental results show that our DSP is more effective than other partitioning schemes, and the proposed architecture outperformed the state-of-the-art results on the public dataset in terms of Area Under the ROC Curve (AUC) and Equal-Error-Rate (EER) by 14.4 % and 28.9%, respectively.