Recent advances in convolutional neural networks

We present SpecAugment, a simple data augmentation method for speech recognition. SpecAugment is applied directly to the feature inputs of a neural network (i.e., filter bank coefficients). The augmentation policy consists of warping the features, masking blocks of frequency channels, and masking blocks of time steps. We apply SpecAugment on Listen, Attend and Spell networks for end-to-end speech recognition tasks. We achieve state-of-the-art performance on the LibriSpeech 960h and Swichboard 300h tasks, outperforming all prior work. On LibriSpeech, we achieve 6.8% WER on test-other without the use of a language model, and 5.8% WER with shallow fusion with a language model. This compares to the previous state-of-the-art hybrid system of 7.5% WER. For Switchboard, we achieve 7.2%/14.6% on the Switchboard/CallHome portion of the Hub5'00 test set without the use of a language model, and 6.8%/14.1% with shallow fusion, which compares to the previous state-of-the-art hybrid system at 8.3%/17.3% WER.

SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition

Thank you very much for downloading applied longitudinal data analysis modeling change and event occurrence. As you may know, people have look hundreds times for their favorite novels like this applied longitudinal data analysis modeling change and event occurrence, but end up in malicious downloads. Rather than enjoying a good book with a cup of coffee in the afternoon, instead they are facing with some infectious virus inside their computer.

Applied Longitudinal Data Analysis Modeling Change And Event Occurrence

OUTPU T 29 OUTPU T 30 OUTPU T 31 OUTPU T 32 OUTPU T 25 OUTPU T 26 OUTPU T 27 OUTPU T 28 OUTPU T 21 OUTPU T 22 OUTPU T 23 OUTPU T 24 OUTPU T 17 OUTPU T 18 OUTPU T 19 OUTPU T 20 OUTPU T 13 OUTPU T 14 OUTPU T 15 OUTPU T 16 OUTPU T 9 OUTPU T 10 OUTPU T 11 OUTPU T 12 OUTPU T 5 OUTPU T 6 OUTPU T 7 OUTPU T 8 OUTPU T 1 OUTPU T 2 OUTPU T 3 OUTPU T 4 29 30 31 32 25 26 27 28 21 22 23 24 17 18 19 20 13 14 15 16 9

A User’s Guide

Recurrent sequence generators conditioned on input data through an attention mechanism have recently shown very good performance on a range of tasks including machine translation, handwriting synthesis [1,2] and image caption generation [3]. We extend the attention-mechanism with features needed for speech recognition. We show that while an adaptation of the model used for machine translation in [2] reaches a competitive 18.7% phoneme error rate (PER) on the TIMET phoneme recognition task, it can only be applied to utterances which are roughly as long as the ones it was trained on. We offer a qualitative explanation of this failure and propose a novel and generic method of adding location-awareness to the attention mechanism to alleviate this issue. The new method yields a model that is robust to long inputs and achieves 18% PER in single utterances and 20% in 10-times longer (repeated) utterances. Finally, we propose a change to the attention mechanism that prevents it from concentrating too much on single frames, which further reduces PER to 17.6% level.

https://proceedings.neurips.cc/paper/2015/file/1068c6e4c8051cfd4e9ea8072e3189e2-Paper.pdf

Attention-based models for speech recognition

Background Even today the reliable diagnosis of the prodromal stages of Alzheimer's disease (AD) remains a great challenge. Our research focuses on the earliest detectable indicators of cognitive decline in mild cognitive impairment (MCI). Since the presence of language impairment has been reported even in the mild stage of AD, the aim of this study is to develop a sensitive neuropsychological screening method which is based on the analysis of spontaneous speech production during performing a memory task. In the future, this can form the basis of an Internet-based interactive screening software for the recognition of MCI. Methods Participants were 38 healthy controls and 48 clinically diagnosed MCI patients. The provoked spontaneous speech by asking the patients to recall the content of 2 short black and white films (one direct, one delayed), and by answering one question. Acoustic parameters (hesitation ratio, speech tempo, length and number of silent and filled pauses, length of utterance) were extracted from the recorded speech signals, first manually (using the Praat software), and then automatically, with an automatic speech recognition (ASR) based tool. First, the extracted parameters were statistically analyzed. Then we applied machine learning algorithms to see whether the MCI and the control group can be discriminated automatically based on the acoustic features. Results The statistical analysis showed significant differences for most of the acoustic parameters (speech tempo, articulation rate, silent pause, hesitation ratio, length of utterance, pause-per-utterance ratio). The most significant differences between the two groups were found in the speech tempo in the delayed recall task, and in the number of pauses for the question-answering task. The fully automated version of the analysis process - that is, using the ASR-based features in combination with machine learning - was able to separate the two classes with an F1-score of 78.8%. Conclusion The temporal analysis of spontaneous speech can be exploited in implementing a new, automatic detection-based tool for screening MCI for the community.

/pdf/a-speech-recognition-based-solution-for-the-automatic-1iup5ze1ze.pdf

A Speech Recognition-based Solution for the Automatic Detection of Mild Cognitive Impairment from Spontaneous Speech.

Identifying Mild Cognitive Impairment and mild Alzheimer’s disease based on spontaneous speech using ASR and linguistic features

Effect of pH on stability and plasmonic properties of cysteine-functionalized silver nanoparticle dispersion

Convolutional neural networks have proved very successful in image recognition, thanks to their tolerance to small translations. They have recently been applied to speech recognition as well, using a spectral representation as input. However, in this case the translations along the two axes - time and frequency - should be handled quite differently. So far, most authors have focused on convolution along the frequency axis, which offers invariance to speaker and speaking style variations. Other researchers have developed a different network architecture that applies time-domain convolution in order to process a longer time-span of input in a hierarchical manner. These two approaches have different background motivations, and both offer significant gains over a standard fully connected network. Here we show that the two network architectures can be readily combined, like their advantages. With the combined model we report an error rate of 16.7% on the TIMIT phone recognition task, a new record on this dataset.

/pdf/combining-time-and-frequency-domain-convolution-in-2i9xvvojfz.pdf

Combining time- and frequency-domain convolution in convolutional neural network-based phone recognition

Deep convolutional neural networks (CNNs) have recently been shown to outperform fully connected deep neural networks (DNNs) both on low-resource and on large-scale speech tasks. Experiments indicate that convolutional networks can attain a 10–15 % relative improvement in the word error rate of large vocabulary recognition tasks over fully connected deep networks. Here, we explore some refinements to CNNs that have not been pursued by other authors. First, the CNN papers published up till now used sigmoid or rectified linear (ReLU) neurons. We will experiment with the maxout activation function proposed recently, which has been shown to outperform the rectifier activation function in fully connected DNNs. We will show that the pooling operation of CNNs and the maxout function are closely related, and so the two technologies can be readily combined to build convolutional maxout networks. Second, we propose to turn the CNN into a hierarchical model. The origins of this approach go back to the era of shallow nets, where the idea of stacking two networks on each other was relatively well known. We will extend this method by fusing the two networks into one joint deep model with many hidden layers and a special structure. We will show that with the hierarchical modelling approach, we can reduce the error rate of the network on an expanded context of input. In the experiments on the Texas Instruments Massachusetts Institute of Technology (TIMIT) phone recognition task, we find that a CNN built from maxout units yields a relative phone error rate reduction of about 4.3 % over ReLU CNNs. Applying the hierarchical modelling scheme to this CNN results in a further relative phone error rate reduction of 5.5 %. Using dropout training, the lowest error rate we get on TIMIT is 16.5 %, which is currently the best result. Besides experimenting on TIMIT, we also evaluate our best models on a low-resource large vocabulary task, and we find that all the proposed modelling improvements give consistently better results for this larger database as well.

/pdf/phone-recognition-with-hierarchical-convolutional-deep-3kw36ad3ml.pdf

László Tóth

Papers

A Speech Recognition-based Solution for the Automatic Detection of Mild Cognitive Impairment from Spontaneous Speech.

Identifying Mild Cognitive Impairment and mild Alzheimer’s disease based on spontaneous speech using ASR and linguistic features

Effect of pH on stability and plasmonic properties of cysteine-functionalized silver nanoparticle dispersion

Combining time- and frequency-domain convolution in convolutional neural network-based phone recognition

Phone recognition with hierarchical convolutional deep maxout networks