How to implement speech synthesis?

Question

Mathematics

I wanted to do an interesting project in the summer and settled on speech synthesis. At the output, I hope to get something at the level of google-translate bot. First without neural networks, and then with the use of deep learning. What can you advise?

Answer 1

It is better to immediately do it qualitatively with neural networks.
Learn Python and NS.
Use the libraries you need and write code.
Package from NVIDIA: OpenSeq2Seq

machine translation (GNMT, Transformer, ConvS2S, …)
speech recognition (DeepSpeech2, Wave2Letter, Jasper, …)
speech commands (RN-50, Jasper)
speech synthesis (Tacotron2, WaveNet…)
language model (LSTM, …)
sentiment analysis (SST ) , IMDB, …)
image classification (ResNet-50)

Voice acting (TTS): Tacotron 2
Lectures:
Lectures 1
Lectures 2
PS:
Synthesis and translation are two different things.
Synthesis is text to sound.
Translation is from one spoken language to another.

Answer 2

Read the source code of the files here: https://github.com/Olga-Yakovleva/RHVoice

Answer 3

First, study the theory, then look at the implementation methods, then write the algorithm.
For example, you can play individual letters, or phonemes, or words, or phrases. You can write it all down ahead of time or generate a robocall just like on your phone. And much more.
Now you are doing something like: "give me a link to the Google api, I will reproduce the text."