Rayhane-mamah / Tacotron-2

DeepMind's Tacotron-2 Tensorflow implementation
MIT License
2.28k stars 905 forks source link

building Commercial synthesizer #247

Open nimamoradi opened 6 years ago

nimamoradi commented 6 years ago

i want to build a commercial synthesizer, in i want to know how much it take to make 1 min audio with gtx1080 gpu cause running python synthesize.py --model='Tacotron-2' take too much time, and also is there a demo server to know how to write REST API for Tacotron

gloriouskilka commented 6 years ago

Hello, @nimamoradi

Did you look at the code of the project you want to build commercical synthesizer on, before creating this issue?

nimamoradi commented 6 years ago

i look at some part of code, and i want make my synthesizer on this project

gloriouskilka commented 6 years ago

There is no HTTP server included in this project. Well, you can start from synthesize.py and write your code with any API

PS you might want to include this fix which is not in master branch yet https://github.com/Rayhane-mamah/Tacotron-2/pull/237

nimamoradi commented 6 years ago

i can write HTTP server my self, problem is how much time it take to synthesize a 50 words for example, because some old wave net based models and also this project synthesize code take too much time

Thien223 commented 6 years ago

I'm working on a Korean language project, my synthesizer take only 2-3 minutes for 18 sentences.

nimamoradi commented 6 years ago

@tdplaza it sound good, how long is audio output and what machine you are working on

Thien223 commented 6 years ago

18 sentences take about 1 minutes of speech. I'm using GTX 1080Ti