Waveform-based speech recognition using hidden filter models: parameter selection and sensitivity to power normalization

By H. Sheikhzadeh, Li Deng. 1994

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Li Deng
2. Dong Yu
3. Jinyu Li
4. Yoshua Bengio
5. Jianfeng Gao
6. Xiaodong He
7. Kyunghyun Cho
8. Kaisheng Yao

Citations in the corpus

• Deep Learning: Methods and Applications (2014, methodology): “The work also showed the importance of raw speech features of spectrogram — back from the long-popular MFCC features toward but not yet reaching the raw speech-waveform level [183, 327].”
• Estimating phoneme class conditional probabilities from raw speech signal using convolutional neural networks (2013, methodology): “In [8], an approach to model raw speech signal was proposed.”
• A tutorial survey of architectures, algorithms, and applications for deep learning (2014, background): “The work also showed the importance of raw speech features of spectrogram – back from the long-popular MFCC features, but not yet reaching the raw speech-waveform level [173, 174].”
• Deep Convex Net: A Scalable Architecture for Speech Pattern Classification (2011, unknown): “An early study for an HMM-like system (i.e., the hidden filter) that models speech waveform directly as the observation can be found in [9].”
• Dynamic Speech Models: Theory, Algorithms, and Applications (2006, background): “The work in [95,96] developed the model having the state-dependent linear prediction performed on the speech waveforms.”
• Recent advances in deep learning for speech research at Microsoft (2013, background): “For speech feature learning and for speech recognition, this goal is condensed to the use of primitive spectral [26]or possibly waveform [46] features.”
• Structured speech modeling (2006, methodology): “• Polynomial trended HMM: — Observable trend functions: This is the simplest trended HMM where there is no uncertainty in polynomial coefficients [ 18 ], [21], [22], [27], [35].”
• An Overview of Deep-Structured Learning for Information Processing (2011, methodology): “…linear units in the hidden layer (Glorot et al., 2011), it is possible to automatically normalize the amplitude variation in the waveform signal, thus overcoming the difficulty encountered in the earlier attempt of using the same raw feature in the HMM based approach (Sheikhzadeh and Deng, 1994).”