Loading...
Please wait, while we are loading the content...
Similar Documents
Real-Time Dereverberation for Deep Neural Network Speech Recognition
| Content Provider | CiteSeerX |
|---|---|
| Author | Schwarz, Andreas Huemmer, Christian Kellermann, Walter |
| Abstract | We evaluate a real-time multi-channel dereverbera-tion method for the application to speech recognition with deep neural networks (DNN). The dereverberation method is based on modeling the reverberated signal as a mixture of a fully coherent direct path signal and a diffuse reverberation component, and estimating the coherent-to-diffuse power ratio (CDR) from the spatial coherence of the signals. The method can operate in real-time, i.e., without requiring processing of entire utterances. We compare CDR estimators which are “blind”, i.e., do not require information about the direction of arrival (DOA) of the target signal, with estimators which make use of a DOA estimate. The impact of the dereverberation method on speech recognition accuracy with different DNN-based acoustic models is investigated with the RE-VERB challenge corpus and the Kaldi speech recognition toolkit. |
| File Format | |
| Access Restriction | Open |
| Subject Keyword | Deep Neural Network Speech Recognition Real-time Dereverberation Dereverberation Method Diffuse Reverberation Component Re-verb Challenge Corpus Spatial Coherence Target Signal Speech Recognition Accuracy Reverberated Signal Cdr Estimator Kaldi Speech Recognition Toolkit Entire Utterance Coherent Direct Path Signal Real-time Multi-channel Dereverbera-tion Method Coherent-to-diffuse Power Ratio Doa Estimate Deep Neural Network Different Dnn-based Acoustic Model |
| Content Type | Text |
