arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1610.03035stat.MLcs.CLcs.LG
收藏

潜序列分解

Latent Sequence Decompositions

Carnegie Mellon University(卡内基梅隆大学);Massachusetts Institute of Technology(麻省理工学院);Google Brain

William Chan, Yu Zhang, Quoc Le, Navdeep Jaitly

摘要:

我们提出了潜在序列分解(LSD)框架。LSD将序列分解为具有作为输入序列和输出序列两者的函数的可变量化输出单元。我们提出了一个训练算法,样本有效的扩展和近似解码算法。我们用华尔街日报的语音识别任务进行实验。我们的LSD模型实现了12.9%的WER,而字符基线为14.8%。当与编码器上的卷积网络相结合时,我们实现了9.6%的WER。

英文摘要:

We present the Latent Sequence Decompositions (LSD) framework. LSD decomposes sequences with variable lengthed output units as a function of both the input sequence and the output sequence. We present a training algorithm which samples valid extensions and an approximate decoding algorithm. We experiment with the Wall Street Journal speech recognition task. Our LSD model achieves 12.9% WER compared to a character baseline of 14.8% WER. When combined with a convolutional network on the encoder, we achieve 9.6% WER.

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。

↑