140623 - Jun Wang
来自cslt Wiki
weekly summary
1\extractor bottleneck features, the dnn structure is 220:1024:1024:40:1024:1024:6950
2\test bottleneck features in ivector framework, the results are posted on the CVSS
next week
1\use pretraining 5:5:5:5:5:1000 for dnn training and re-extractor the bottleneck features.
2\also test in the ivector framework
