低调奋进声明：工作以来主要从事TTS工作，平时看些文章做些笔记。文章中难免存在错误的地方，还望大家海涵。平时搜集一些资料，方便查阅学习：
TTS 论文列表 http://yqli.tech/page/tts_paper.html TTS
开源数据 http://yqli.tech/page/data.html
如转载，请标明出处。
本文整理一下GAN在语音合成声学模型上的研究，统计可能不全。GAN在语音合成中的应用更多是在声码器部分，而在声学模型部分的研究很少。使用GAN的目的更多解决过平滑问题，从而使韵律更好，情感表达更丰富。本文搜集近几年的文章，做简短的总结：
1)TFGAN: A Lightweight Library for Generative Adversarial Networks (2017)
https://ai.googleblog.com/2017/12/tfgan-lightweight-library-for.html
2) A new gan-based end-to-end tts training algorithm (2019)
https://arxiv.org/pdf/1904.04775.pdf
3) A New End-to-End Long-Time Speech Synthesis System Based on Tacotron2 （2019）
https://sci-hub.se/https://doi.org/10.1145/3364908.3365292
非taoctron实验
4) Statistical parametric speech synthesis incorporating generative adversarial networks （2018）
https://sci-hub.se/https://doi.org/10.1109/TASLP.2017.2761547
5) Wasserstein GAN and waveform loss-based acoustic model training for multi-speaker text-to-speech synthesis systems using a WaveNet vocoder （2018）
https://arxiv.org/pdf/1807.11679.pdf
6) Reducing over-smoothness in speech synthesis using Generative Adversarial Networks （2018）
https://arxiv.org/pdf/1810.10989.pdf
7) High-quality Speech Synthesis Using Super-resolution Mel-Spectrogram（2019）
https://arxiv.org/pdf/1912.01167.pdf
8) WGANSing: A Multi-Voice Singing Voice Synthesizer Based on the Wasserstein-GAN （2019）
https://arxiv.org/pdf/1903.10729.pdf
9) HiFiSinger: Towards High-Fidelity Neural Singing Voice Synthesis (2020)
https://arxiv.org/pdf/2009.01776.pdf
10) GAN acoustic model for Kazakh speech synthesis（2021）
基于tacotron的实验主要在tacotron系统上进行的优化。
1 TFGAN: A Lightweight Library for Generative Adversarial Networks (2017)
3 A New End-to-End Long-Time Speech Synthesis System Based on Tacotron2 （2019）
添加了三个分辨器：prosody, sound quality和alignments discriminator。tacotron2 删除mse损失函数而使用ganloss。
6 Reducing over-smoothness in speech synthesis using Generative Adversarial Networks （2018）
7 High-quality Speech Synthesis Using Super-resolution Mel-Spectrogram（2019）
今年的文章，没找到免费的pdf，没钱下载，读者有渠道下载后可以分享我一份。
