|
Here you can download some of the British English acoustic models I trained using the WSJCAM0 corpus. I don't have a training recipe for these.
These use the 40 phone set from the CMU dictionary (without stress) and were trained using 1s_12c_12d_3p_12dd acoustic features (continuous models) or s2_4x features (semi-continuous models).
Sampling rate |
Adapted to |
Type |
Topology |
Senoes |
Gaussians |
Size |
Zip file |
16kHz |
- |
cont |
3 states, no skips |
3000 |
8 |
8MB |
Download |
16kHz |
Male, MLLR-40 + MAP |
cont |
3 states, no skips |
3000 |
8 |
8MB |
Download |
16kHz |
Female, MLLR-40 + MAP |
cont |
3 states, no skips |
3000 |
8 |
8MB |
Download |
16kHz |
- |
semi |
5 states, skips |
4000 |
256 |
16MB |
Download |
16kHz |
Male, MLLR-1 + MAP |
semi |
5 states, skips |
4000 |
256 |
7MB |
Download |
16kHz |
Female, MLLR-1 + MAP |
semi |
5 states, skips |
4000 |
256 |
7MB |
Download |
8kHz |
- |
semi |
5 states, skips |
4000 |
256 |
16MB |
Download |
8kHz |
Male, MLLR-1 + MAP |
semi |
5 states, skips |
4000 |
256 |
9MB |
Download |
8kHz |
Female, MLLR-1 + MAP |
semi |
5 states, skips |
4000 |
256 |
9MB |
Download |
|