九色国产,午夜在线视频,新黄色网址,九九色综合,天天做夜夜做久久做狠狠,天天躁夜夜躁狠狠躁2021a,久久不卡一区二区三区

Seq2Seq

參考論文：Listen, Attend and Spell: A Neural Network for Large Vocabulary Conversational Speech Recognition. ICASSP 2016（William Chan, Navdeep Jaitly, Quoc V. Le, Oriol Vinyals）

CTC

參考論文：Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks. ICML 2006（AlexGraves, SantiagoFernández,FaustinoGomez）

這里 A 是一條合法的 x 和 y 的對(duì)應(yīng)路徑，a_t 代表 t 時(shí)刻 X 所對(duì)應(yīng)的輸出。

了解更多的推導(dǎo)細(xì)節(jié)： https://distill.pub/2017/ctc/ 

Transducer

參考論文：Sequence Transduction with Recurrent Neural Networks. arXiv 2012（Alex Graves）

數(shù)據(jù)增強(qiáng)

參考論文：SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition. INTERSPEECH 2019（Daniel S. Park, William Chan, Yu Zhang, Chung-Cheng Chiu, Barret Zoph, Ekin D. Cubuk, Quoc V. Le）

延遲優(yōu)化

參考論文：Towards Fast and Accurate Streaming End-to-End ASR. ICCASP 2019（Bo Li, Shuo-yiin Chang, Tara N. Sainath, Ruoming Pang, Yanzhang He, Trevor Strohman, Yonghui Wu）

參考論文：On the Comparison of Popular End-to-End Models for Large Scale Speech Recognition. InterSpeech 2020（Jinyu Li, Yu Wu, Yashesh Gaur, Chengyi Wang, Rui Zhao, Shujie Liu）