Experiments with noisy data
Enhancement methods:
OMLSA: baseline speech enhancement method
RNN-V: RNN enhancement of Vocoded parameters
RNN-DFT: RNN enhancement of DFT based parameters
Natural and vocoded speech samples:
| |
FEMALE |
MALE |
| |
Bus 7.5dB |
Cafe 12.5dB |
Bus 7.5dB |
Cafe 12.5dB |
| |
Natural |
Vocoded |
Natural |
Vocoded |
Natural |
Vocoded |
Natural |
Vocoded |
| CLEAN |
|
|
|
|
|
|
|
|
| NOISY |
|
|
|
|
|
|
|
|
| OMLSA |
|
|
|
|
|
|
|
|
| RNN-V |
X |
|
X |
|
X |
|
X |
|
| RNN-DFT |
|
|
|
|
|
|
|
|
HMM-based synthetic speech samples:
| |
FEMALE |
MALE |
| |
Sample 1 |
Sample 2 |
Sample 3 |
Sample 4 |
Sample 1 |
Sample 2 |
Sample 3 |
Sample 4 |
| CLEAN |
|
|
|
|
|
|
|
|
| NOISY |
|
|
|
|
|
|
|
|
| OMLSA |
|
|
|
|
|
|
|
|
| RNN-V |
|
|
|
|
|
|
|
|
| RNN-DFT |
|
|
|
|
|
|
|
|