Kazuki Yamauchi, Wataru Nakata, Yuki Saito, Hiroshi Saruwatari
The University of Tokyo, Japan.
Demo page
Compared methods
We present samples of synthesized speech using the following decoding strategies.
- ・Greedy decoding
- ・Naive sampling
- ・Top-k top-p sampling
- ・Sequence-wise BOK-PRP (proposed)
- ・Block-wise BOK-PRP (proposed)
Samples of synthetic speech
| Greedy decoding |
|
|
|
| Naive sampling |
|
|
|
| Top-k top-p sampling |
|
|
|
| Sequence-wise BOK-PRP (proposed) |
|
|
|
| Block-wise BOK-PRP (proposed) |
|
|
|
| Ground truth |
|
|
|