人工智能培训

搜索

深度学习论文:使用WAVENET自动编码器进行无监督语音表示学习(Unsupervised speech representation learning usi

[复制链接]
usjerseys889 发表于 2019-1-28 11:23:05 | 显示全部楼层 |阅读模式
usjerseys889 2019-1-28 11:23:05 249 0 显示全部楼层
深度学习论文:使用WAVENET自动编码器进行无监督语音表示学习(Unsupervised speech representation learning using WaveNet autoencoders)我们通过将自动编码神经网络应用于语音波形来考虑无监督提取有意义的语音潜在表示的任务。目标是学习能够从信号中捕获高级语义内容的表示,例如,音素身份,同时不会混淆信号中的低级细节,例如底层音高轮廓或背景噪音。自动编码器模型的行为取决于应用于潜在表示的约束类型。我们比较了三种变体:简单的降维瓶颈,高斯变分自动编码器(VAE)和离散矢量量化VAE(VQ-VAE)。我们根据说话人的独立性,预测语音内容的能力以及精确重建单个谱图帧的能力来分析学习表征的质量。此外,对于使用VQ-VAE提取的差异编码,我们测量将它们映射到电话的容易程度。我们引入了一种正则化方案,该方案强制表示集中于话语的语音内容,并报告性能与ZeroSpeech 2017无监督声学单元发现任务中的顶级条目相当。
We consider the task of unsupervised extraction of meaningful latentrepresentations of speech by applying autoencoding neural networks to speechwaveforms.The goal is to learn a representation able to capture high levelsemantic content from the signal, e.g.phoneme identities, while beinginvariant to confounding low level details in the signal such as the underlyingpitch contour or background noise.The behavior of autoencoder models dependson the kind of constraint that is applied to the latent representation.Wecompare three variants: a simple dimensionality reduction bottleneck, aGaussian Variational Autoencoder (VAE), and a discrete Vector Quantized VAE(VQ-VAE).We analyze the quality of learned representations in terms of speakerindependence, the ability to predict phonetic content, and the ability toaccurately reconstruct individual spectrogram frames.Moreover, for discreteencodings extracted using the VQ-VAE, we measure the ease of mapping them tophonemes.We introduce a regularization scheme that forces the representationsto focus on the phonetic content of the utterance and report performancecomparable with the top entries in the ZeroSpeech 2017 unsupervised acousticunit discovery task.深度学习论文:使用WAVENET自动编码器进行无监督语音表示学习(Unsupervised speech representation learning using WaveNet autoencoders) lYqQsV2sQass4D7s.jpg
URL地址:https://arxiv.org/abs/1901.08810     ----pdf下载地址:https://arxiv.org/pdf/1901.08810    ----深度学习论文:使用WAVENET自动编码器进行无监督语音表示学习(Unsupervised speech representation learning using WaveNet autoencoders)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

usjerseys889当前离线
新手上路

查看:249 | 回复:0

快速回复 返回顶部 返回列表