Speech synthesis: tacotron explains [end-to-end speech synthesis model] [compared with traditional speech synthesis, it does not have complex phonetics and acoustic feature modules, but only uses ＜ te

​ ​ Tacotron The model is the first real end-to-end TTS Deep neural network model . Compared with traditional speech synthesis , It has no complex phonetic and acoustic feature modules , But just < Text sequence , Speech spectrum > The neural network is trained by pairing data sets , So it simplifies a lot of processes . then Tacotron Use Griffin-Lim The algorithm estimates the phase of the amplitude spectrum predicted by the network , And then a short-time Fourier （Short-Time Fourier Transform,STFT） inverse transformation , Realize the function of end-to-end speech synthesis .Tacotron The overall architecture of is shown in the figure below ：

---

---

---

Reference material ：
Tacotron as well as Tacotron2 Detailed explanation
speech synthesis ( One )：Tacotron