超单调对齐搜索

arXiv - EE - Audio and Speech Processing Pub Date : 2024-09-12 DOI:arxiv-2409.07704

Junhyeok Lee, Hyeongju Kim

{"title":"超单调对齐搜索","authors":"Junhyeok Lee, Hyeongju Kim","doi":"arxiv-2409.07704","DOIUrl":null,"url":null,"abstract":"Monotonic alignment search (MAS), introduced by Glow-TTS, is one of the most\npopular algorithm in TTS to estimate unknown alignments between text and\nspeech. Since this algorithm needs to search for the most probable alignment\nwith dynamic programming by caching all paths, the time complexity of the\nalgorithm is $O(T \\times S)$. The authors of Glow-TTS run this algorithm on\nCPU, and while they mentioned it is difficult to parallelize, we found that MAS\ncan be parallelized in text-length dimension and CPU execution consumes an\ninordinate amount of time for inter-device copy. Therefore, we implemented a\nTriton kernel and PyTorch JIT script to accelerate MAS on GPU without\ninter-device copy. As a result, Super-MAS Triton kernel is up to 72 times\nfaster in the extreme-length case. The code is available at\n\\url{https://github.com/supertone-inc/super-monotonic-align}.","PeriodicalId":501284,"journal":{"name":"arXiv - EE - Audio and Speech Processing","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-09-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Super Monotonic Alignment Search\",\"authors\":\"Junhyeok Lee, Hyeongju Kim\",\"doi\":\"arxiv-2409.07704\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Monotonic alignment search (MAS), introduced by Glow-TTS, is one of the most\\npopular algorithm in TTS to estimate unknown alignments between text and\\nspeech. Since this algorithm needs to search for the most probable alignment\\nwith dynamic programming by caching all paths, the time complexity of the\\nalgorithm is $O(T \\\\times S)$. The authors of Glow-TTS run this algorithm on\\nCPU, and while they mentioned it is difficult to parallelize, we found that MAS\\ncan be parallelized in text-length dimension and CPU execution consumes an\\ninordinate amount of time for inter-device copy. Therefore, we implemented a\\nTriton kernel and PyTorch JIT script to accelerate MAS on GPU without\\ninter-device copy. As a result, Super-MAS Triton kernel is up to 72 times\\nfaster in the extreme-length case. The code is available at\\n\\\\url{https://github.com/supertone-inc/super-monotonic-align}.\",\"PeriodicalId\":501284,\"journal\":{\"name\":\"arXiv - EE - Audio and Speech Processing\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-12\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - EE - Audio and Speech Processing\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.07704\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - EE - Audio and Speech Processing","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.07704","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}

引用次数: 0

摘要

单调对齐搜索（Monotonic alignment search，MAS）由Glow-TTS引入，是TTS中最流行的算法之一，用于估计文本和语音之间的未知对齐。由于该算法需要通过缓存所有路径，以动态编程的方式搜索最可能的对齐方式，因此该算法的时间复杂度为 $O(T\times S)$。Glow-TTS的作者在CPU上运行了该算法，虽然他们提到该算法很难并行化，但我们发现MAS可以在文本长度维度上并行化，CPU执行会消耗过多的时间用于设备间复制。因此，我们采用了Triton内核和PyTorch JIT脚本来加速GPU上的MAS，而无需进行设备间拷贝。结果，Super-MAS Triton 内核在极端长度情况下的速度提高了 72 倍。代码可在（url{https://github.com/supertone-inc/super-monotonic-align}.

本文章由计算机程序翻译，如有差异，请以英文原文为准。

查看原文本刊更多论文

Super Monotonic Alignment Search

Monotonic alignment search (MAS), introduced by Glow-TTS, is one of the most popular algorithm in TTS to estimate unknown alignments between text and speech. Since this algorithm needs to search for the most probable alignment with dynamic programming by caching all paths, the time complexity of the algorithm is $O(T \times S)$. The authors of Glow-TTS run this algorithm on CPU, and while they mentioned it is difficult to parallelize, we found that MAS can be parallelized in text-length dimension and CPU execution consumes an inordinate amount of time for inter-device copy. Therefore, we implemented a Triton kernel and PyTorch JIT script to accelerate MAS on GPU without inter-device copy. As a result, Super-MAS Triton kernel is up to 72 times faster in the extreme-length case. The code is available at \url{https://github.com/supertone-inc/super-monotonic-align}.

求助全文

通过发布文献求助，成功后即可免费获取论文全文。去求助

来源期刊

arXiv - EE - Audio and Speech Processing

自引率

0.00%

发文量