Impact of deep MLP architecture on different acoustic modeling techniques for under-resourced speech recognition
By David Imseng, Petr Motlícek, Philip N. Garner, Hervé Bourlard. 2013
31 citations recorded by the source corpus.
Posterior based acoustic modeling techniques such as Kullback-Leibler divergence based HMM (KL-HMM) and Tandem are able to exploit out-of-language data through posterior features, estimated by a Multi-Layer Perceptron (MLP). In this paper, we investigate the performance of posterior based approaches in the context of under-resourced speech recognition when a standard three-layer MLP is replaced by a deeper five-layer MLP. The deeper MLP architecture yields similar gains of about 15% (relative) for Tandem, KL-HMM as well as for a hybrid HMM/MLP system that directly uses the posterior estimates as emission probabilities. The best performing system, a bilingual KL-HMM based on a deep MLP, jointly trained on Afrikaans and Dutch data, performs 13% better than a hybrid system using the same bilingual MLP and 26% better than a subspace Gaussian mixture system only trained on Afrikaans data.
Subfields: Artificial Intelligence, Signal Processing
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Hervé Bourlard
- Petr Motlícek
- David Imseng
- Ngoc Thang Vu
- Tanja Schultz
- Daniel Povey
- Li Deng
- Sanjeev Khudanpur
In Signal Processing
- Hervé Bourlard
- Petr Motlícek
- David Imseng
- Tanja Schultz
- Daniel Povey
- Li Deng
- Sanjeev Khudanpur
- Dong Yu
Citations in the corpus
- Multilingual deep neural network based acoustic modeling for rapid language adaptation (2014, methodology): “However, in [14], only DNNs with three hidden layers were used, pre-training was not applied and the setup was bilingual (Afrikaans and Dutch).”