These are a series of projects that we are working on with an aim to building high quality speech synthesis systems.
Updates
In December 2017, we have tried experiments on upgrading our speech representation to WORLD.
As of March 2018, festvox voice building tools support WORLD as representation.
Checkout our repo and notebook
In Summer 2018, we are working on upgrading our vocoder to WaveNet.
21 December 2017
FESTIVAL + WORLD + 6 layer DNN
AWB ARCTIC
ARCTIC A0029 ABS
ARCTIC A0029 TEST
29 December 2017
FESTIVAL + WORLD + 6 layer DNN
AWB ARCTIC
ARCTIC A0029 128 Tanh 0.2 Dropout SGD
ARCTIC A0029 128 Tanh 0.3 Dropout ADAM
ARCTIC A0029 200 Tanh 0.2 Dropout ADAM
ARCTIC A0029 512 Tanh 0.1 Dropout ADAM
ARCTIC A0029 512 Tanh 0.3 Dropout ADAM
ARCTIC A0029 1024 Tanh 0.1 Dropout ADAM
ARCTIC A0029 1024 Tanh 0.2 Dropout ADAM
ARCTIC A0029 1024 Tanh 0.3 Dropout ADAM
ARCTIC A0029 TEST (previous week)
05 January 2018
FESTIVAL + WORLD + 6 layer SELU DNN
AWB ARCTIC
ARCTIC A0029 512 SGD No Dropout No Context No Normalization
ARCTIC A0039 512 SGD No Dropout No Context No Normalization
12 January 2018
FESTIVAL + WORLD + 6 layer SELU DNN + Dynet
AWB ARCTIC
ARCTIC A0029 1024 SGD No Dropout No Context No Normalization
ARCTIC A0029 1024 SGD 0.2 Dropout No Context No Normalization