Notes:
- Source and target speaker's speech was resynthesized from NMF decomposition
- Row 1, 2, and 4 use 300 exemplars selected using the proposed FWD algorithm
- Row 3 uses 204 exemplars selected using the proposed BKW algorithm
- Please refer to our paper for more details
| Source speaker | Target speaker | Voice conversion |
|---|---|---|
|
BDL (male) |
RMS (male) |
|
|
BDL (male) |
SLT (female) |
|
|
CLB (female) |
RMS (male) |
|
|
CLB (female) |
SLT (female) |