Jurafsky & Martin is the canonical textbook but it's pre-deep learning. Check out the wavenet paper[0], and Alan Black's Festvox[1].

[0] https://deepmind.com/blog/wavenet-generative-model-raw-audio... [1] http://www.festvox.org/

These new models are replacing almost everything that's been developed for TTS in the last 30 years with neural nets. So you don't need to study TTS anymore. Study neural nets and then read this paper and you will understand the state of the art in TTS.

This is complete nonsense. You need to understand the problem you are addressing even if you plan to use deep learning methods for them.

Saying you don't need to know anything about TTS is like developing a self-driving car and saying you don't need to know anything about the rules of the road. For one thing, how are you supposed to know when you got it wrong? And what things it is important to get right?

I've used AT&T's Natural Voices SDK about 10-15 years ago? It allowed you to create your own voices, but it's a hell of a lot of work.

this book is a classic in the field :

http://svr-www.eng.cam.ac.uk/~pat40/ttsbook_draft_2.pdf