In industry, automatic speech recognition has come to be a competitive feature for embedded products with poor hardware resources. In this work, we propose a tiny end-to-end speech recognition model that is lightweight and easily deployable on edge platforms. First, instead of sophisticated network structures, such as recurrent neural networks, transformers, etc., the model we propose mainly uses convolutional neural networks as its backbone. This ensures that our model is supported by most software development kits for embedded devices. Second, we adopt the basic unit of MobileNet-v3, which performs well in computer vision tasks, and integrate the features of the hidden layer at different scales, thus compressing the number of parameters of the model to less than 1 M and achieving an accuracy greater than that of some traditional models. Third, in order to further reduce the CPU computation, we directly extract acoustic representations from 1-dimensional speech waveforms and use a self-supervised learning approach to encourage the convergence of the model. Finally, to solve some problems where hardware resources are relatively weak, we use a prefix beam search decoder to dynamically extend the search path with an optimized pruning strategy and an additional initialism language model to capture the probability of between-words in advance and thus avoid premature pruning of correct words. In our experiments, according to a number of evaluation categories, our end-to-end model outperformed several tiny speech recognition models used for embedded devices in related work.

In this work, we propose a new automatic speech recognition (ASR) system based on feature learning and an end-to-end training procedure for air traffic control (ATC) systems. The proposed model integrates the feature learning block, recurrent neural network (RNN), and connectionist temporal classification loss to build an end-to-end ASR model. Facing the complex environments of ATC speech, instead of the handcrafted features, a learning block is designed to extract informative features from raw waveforms for acoustic modeling. Both the SincNet and 1D convolution blocks are applied to process the raw waveforms, whose outputs are concatenated to the RNN layers for the temporal modeling. Thanks to the ability to learn representations from raw waveforms, the proposed model can be optimized in a complete end-to-end manner, i.e., from waveform to text. Finally, the multilingual issue in the ATC domain is also considered to achieve the ASR task by constructing a combined vocabulary of Chinese characters and English letters. The proposed approach is validated on a multilingual real-world corpus (ATCSpeech), and the experimental results demonstrate that the proposed approach outperforms other baselines, achieving a 6.9% character error rate.

Recently, robust speech recognition for real-world applications has attracted much attention. This paper proposes a robust speech recognition method based on the teacher-student learning framework for domain adaptation. In particular, the student network will be trained based on a novel optimization criterion defined by the encoder outputs of both teacher and student networks rather than the final output posterior probabilities, which aims to make the noisy audio map to the same embedding space as clean audio, so that the student network is adaptive in the noise domain. Comparative experiments demonstrate that the proposed method obtained good robustness against noise.

Speech recognition is a technique that recognizes words and sentences in audio form and converts them into text sentences. Currently, with the advancement of deep learning technologies, speech recognition has achieved very satisfactory results close to human abilities. However, there are still limitations in identification results such as lack of punctuation, capitalization, and standardized numerical data. Vietnamese also contains local words, homonyms, etc, which make it difficult to read and understand the identification results for users as well as to perform the next tasks in Natural Language Processing (NLP). In this paper, we propose to combine the transformer decoder with conditional random field (CRF) to restore punctuation and capitalization for the Vietnamese automatic speech recognition (ASR) output. By chunking input sentences and merging output sequences, it is possible to handle longer strings with greater accuracy. Experiments show that the method proposed in the Vietnamese post-speech recognition dataset delivers the best results.

This paper demonstrates latent word recurrent neural network language models (LW-RNN-LMs) for enhancing automatic speech recognition (ASR). LW-RNN-LMs are constructed so as to pick up advantages in both recurrent neural network language models (RNN-LMs) and latent word language models (LW-LMs). The RNN-LMs can capture long-range context information and offer strong performance, and the LW-LMs are robust for out-of-domain tasks based on the latent word space modeling. However, the RNN-LMs cannot explicitly capture hidden relationships behind observed words since a concept of a latent variable space is not present. In addition, the LW-LMs cannot take into account long-range relationships between latent words. Our idea is to combine RNN-LM and LW-LM so as to compensate individual disadvantages. The LW-RNN-LMs can support both a latent variable space modeling as well as LW-LMs and a long-range relationship modeling as well as RNN-LMs at the same time. From the viewpoint of RNN-LMs, LW-RNN-LM can be considered as a soft class RNN-LM with a vast latent variable space. In contrast, from the viewpoint of LW-LMs, LW-RNN-LM can be considered as an LW-LM that uses the RNN structure for latent variable modeling instead of an n-gram structure. This paper also details a parameter inference method and two kinds of implementation methods, an n-gram approximation and a Viterbi approximation, for introducing the LW-LM to ASR. Our experiments show effectiveness of LW-RNN-LMs on a perplexity evaluation for the Penn Treebank corpus and an ASR evaluation for Japanese spontaneous speech tasks.

This paper proposes a novel domain adaptation method that can utilize out-of-domain text resources and partially domain matched text resources in language modeling. A major problem in domain adaptation is that it is hard to obtain adequate adaptation effects from out-of-domain text resources. To tackle the problem, our idea is to carry out model merger in a latent variable space created from latent words language models (LWLMs). The latent variables in the LWLMs are represented as specific words selected from the observed word space, so LWLMs can share a common latent variable space. It enables us to perform flexible mixture modeling with consideration of the latent variable space. This paper presents two types of mixture modeling, i.e., LWLM mixture models and LWLM cross-mixture models. The LWLM mixture models can perform a latent word space mixture modeling to mitigate domain mismatch problem. Furthermore, in the LWLM cross-mixture models, LMs which individually constructed from partially matched text resources are split into two element models, each of which can be subjected to mixture modeling. For the approaches, this paper also describes methods to optimize mixture weights using a validation data set. Experiments show that the mixture in latent word space can achieve performance improvements for both target domain and out-of-domain compared with that in observed word space.

Previous research has shown that transcripts generated by automatic speech recognition (ASR) technologies can improve the listening comprehension of non-native speakers (NNSs). However, we still lack a detailed understanding of how ASR transcripts affect listening comprehension of NNSs. To explore this issue, we conducted two studies. The first study examined how the current presentation of ASR transcripts impacted NNSs' listening comprehension. 20 NNSs engaged in two listening tasks, each in different conditions: C1) audio only and C2) audio+ASR transcripts. The participants pressed a button whenever they encountered a comprehension problem, and explained each problem in the subsequent interviews. From our data analysis, we found that NNSs adopted different strategies when using the ASR transcripts; some followed the transcripts throughout the listening; some only checked them when necessary. NNSs also appeared to face difficulties following imperfect and slightly delayed transcripts while listening to speech - many reported difficulties concentrating on listening/reading or shifting between the two. The second study explored how different display methods of ASR transcripts affected NNSs' listening experiences. We focused on two display methods: 1) accuracy-oriented display which shows transcripts only after the completion of speech input analysis, and 2) speed-oriented display which shows the interim analysis results of speech input. We conducted a laboratory experiment with 22 NNSs who engaged in two listening tasks with ASR transcripts presented via the two display methods. We found that the more the NNSs paid attention to listening to the audio, the more they tended to prefer the speed-oriented transcripts, and vice versa. Mismatched transcripts were found to have negative effects on NNSs' listening comprehension. Our findings have implications for improving the presentation methods of ASR transcripts to more effectively support NNSs.

This paper aims to improve the domain robustness of language modeling for automatic speech recognition (ASR). To this end, we focus on applying the latent words language model (LWLM) to ASR. LWLMs are generative models whose structure is based on Bayesian soft class-based modeling with vast latent variable space. Their flexible attributes help us to efficiently realize the effects of smoothing and dimensionality reduction and so address the data sparseness problem; LWLMs constructed from limited domain data are expected to robustly cover unknown multiple domains in ASR. However, the attribute flexibility seriously increases computation complexity. If we rigorously compute the generative probability for an observed word sequence, we must consider the huge quantities of all possible latent word assignments. Since this is computationally impractical, some approximation is inevitable for ASR implementation. To solve the problem and apply this approach to ASR, this paper presents an n-gram approximation of LWLM. The n-gram approximation is a method that approximates LWLM as a simple back-off n-gram structure, and offers LWLM-based robust one-pass ASR decoding. Our experiments verify the effectiveness of our approach by evaluating perplexity and ASR performance in not only in-domain data sets but also out-of-domain data sets.

Introducing pronunciation models into decoding has been proven to be benefit to LVCSR. In this paper, a discriminative pronunciation modeling method is presented, within the framework of the Minimum Phone Error (MPE) training for HMM/GMM. In order to bring the pronunciation models into the MPE training, the auxiliary function is rewritten at word level and decomposes into two parts. One is for co-training the acoustic models, and the other is for discriminatively training the pronunciation models. On Mandarin conversational telephone speech recognition task, compared to the baseline using a canonical lexicon, the discriminative pronunciation models reduced the absolute Character Error Rate (CER) by 0.7% on LDC test set, and with the acoustic model co-training, 0.8% additional CER decrease had been achieved.

In this paper, a new database suitable for HMM-based automatic Filipino speech recognition is described for the purpose of training a domain-independent, large-vocabulary continuous speech recognition system. Although it is known that high-performance speech recognition systems depend on a superior speech database used in the training stage, due to the lack of such an appropriate database, previous reports on Filipino speech recognition had to contend with serious data sparsity issues. In this paper we alleviate such sparsity through appropriate data analysis that makes the evaluation results more reliable. The best system is identified through its low word-error rate to a cross-validation set containing almost three hours of unknown speech data. Language-dependent problems are discussed, and their impact on accuracy was analyzed. The approach is currently data driven, however it serves as a competent baseline model for succeeding future developments.

Mel Frequency Cepstral Coefficients (MFCC) are the most popular acoustic features used in automatic speech recognition (ASR), mainly because the coefficients capture the most useful information of the speech and fit well with the assumptions used in hidden Markov models. As is well known, MFCCs already employ several principles which have known counterparts in the peripheral properties of human hearing: decoupling across frequency, mel-warping of the frequency axis, log-compression of energy, etc. It is natural to introduce more mechanisms in the auditory periphery to improve the noise robustness of MFCC. In this paper, a k-nearest neighbors based frequency masking filter is proposed to reduce the audibility of spectra valleys which are sensitive to noise. Besides, Moore and Glasberg's critical band equivalent rectangular bandwidth (ERB) expression is utilized to determine the filter bandwidth. Furthermore, a new bandpass infinite impulse response (IIR) filter is proposed to imitate the temporal masking phenomenon of the human auditory system. These three auditory perceptual mechanisms are combined with the standard MFCC algorithm in order to investigate their effects on ASR performance, and a revised MFCC extraction scheme is presented. Recognition performances with the standard MFCC, RASTA perceptual linear prediction (RASTA-PLP) and the proposed feature extraction scheme are evaluated on a medium-vocabulary isolated-word recognition task and a more complex large vocabulary continuous speech recognition (LVCSR) task. Experimental results show that consistent robustness against background noise is achieved on these two tasks, and the proposed method outperforms both the standard MFCC and RASTA-PLP.

This paper presents our recent work in regard to building Large Vocabulary Continuous Speech Recognition (LVCSR) systems for the Thai, Indonesian, and Chinese languages. For Thai, since there is no word boundary in the written form, we have proposed a new method for automatically creating word-like units from a text corpus, and applied topic and speaking style adaptation to the language model to recognize spoken-style utterances. For Indonesian, we have applied proper noun-specific adaptation to acoustic modeling, and rule-based English-to-Indonesian phoneme mapping to solve the problem of large variation in proper noun and English word pronunciation in a spoken-query information retrieval system. In spoken Chinese, long organization names are frequently abbreviated, and abbreviated utterances cannot be recognized if the abbreviations are not included in the dictionary. We have proposed a new method for automatically generating Chinese abbreviations, and by expanding the vocabulary using the generated abbreviations, we have significantly improved the performance of spoken query-based search.

This paper presents a novel method for improving the readability of automatic speech recognition (ASR) results for classroom lectures. Because speech in a classroom is spontaneous and contains many ill-formed utterances with various disfluencies, the ASR result should be edited to improve the readability before presenting it to users, by applying some operations such as removing disfluencies, determining sentence boundaries, inserting punctuation marks and repairing dropped words. Owing to the presence of many kinds of domain-dependent words and casual styles, even state-of-the-art recognizers can only achieve a 30-50% word error rate for speech in classroom lectures. Therefore, a method for improving the readability of ASR results is needed to make it robust to recognition errors. We can use multiple hypotheses instead of the single-best hypothesis as a method to achieve a robust response to recognition errors. However, if the multiple hypotheses are represented by a lattice (or a confusion network), it is difficult to utilize sentence-level knowledge, such as chunking and dependency parsing, which are imperative for determining the discourse structure and therefore imperative for improving readability. In this paper, we propose a novel algorithm that infers clean, readable transcripts from spontaneous multiple hypotheses represented by a confusion network while integrating sentence-level knowledge. Automatic and manual evaluations showed that using multiple hypotheses and sentence-level knowledge is effective to improve the readability of ASR results, while preserving the understandability.

We propose a novel scheme to learn a language model (LM) for automatic speech recognition (ASR) directly from continuous speech. In the proposed method, we first generate phoneme lattices using an acoustic model with no linguistic constraints, then perform training over these phoneme lattices, simultaneously learning both lexical units and an LM. As a statistical framework for this learning problem, we use non-parametric Bayesian statistics, which make it possible to balance the learned model's complexity (such as the size of the learned vocabulary) and expressive power, and provide a principled learning algorithm through the use of Gibbs sampling. Implementation is performed using weighted finite state transducers (WFSTs), which allow for the simple handling of lattice input. Experimental results on natural, adult-directed speech demonstrate that LMs built using only continuous speech are able to significantly reduce ASR phoneme error rates. The proposed technique of joint Bayesian learning of lexical units and an LM over lattices is shown to significantly contribute to this improvement.

In this paper, we propose a PLSA-based language model for sports-related live speech. This model is implemented using a unigram rescaling technique that combines a topic model and an n-gram. In the conventional method, unigram rescaling is performed with a topic distribution estimated from a recognized transcription history. This method can improve the performance, but it cannot express topic transition. By incorporating the concept of topic transition, it is expected that the recognition performance will be improved. Thus, the proposed method employs a "Topic HMM" instead of a history to estimate the topic distribution. The Topic HMM is an Ergodic HMM that expresses typical topic distributions as well as topic transition probabilities. Word accuracy results from our experiments confirmed the superiority of the proposed method over a trigram and a PLSA-based conventional method that uses a recognized history.

In this paper, signal processing techniques which can be applied to automatic speech recognition to improve its robustness are reviewed. The choice of signal processing techniques is strongly dependent on the scenario of the applications. The analysis of scenario and the choice of suitable signal processing techniques are shown through two examples.

This paper describes a robust automatic speech recognition (ASR) system with less computation. Acoustic models of a hidden Markov model (HMM)-based classifier include various types of hidden factors such as speaker-specific characteristics, coarticulation, and an acoustic environment, etc. If there exists a canonicalization process that can recover the degraded margin of acoustic likelihoods between correct phonemes and other ones caused by hidden factors, the robustness of ASR systems can be improved. In this paper, we introduce a canonicalization method that is composed of multiple distinctive phonetic feature (DPF) extractors corresponding to each hidden factor canonicalization, and a DPF selector which selects an optimum DPF vector as an input of the HMM-based classifier. The proposed method resolves gender factors and speaker variability, and eliminates noise factors by applying the canonicalzation based on the DPF extractors and two-stage Wiener filtering. In the experiment on AURORA-2J, the proposed method provides higher word accuracy under clean training and significant improvement of word accuracy in low signal-to-noise ratio (SNR) under multi-condition training compared to a standard ASR system with mel frequency ceptral coeffient (MFCC) parameters. Moreover, the proposed method requires a reduced, two-fifth, Gaussian mixture components and less memory to achieve accurate ASR.

The open-vocabulary name recognition technique is one of the most challenging tasks in the application of automatic Chinese speech recognition technology. It can be used as the free name input method for telephony speech applications and automatic directory assistance systems. A Chinese name usually has two to three characters, each of which is pronounced as a single tonal syllable. Obviously, it is very confusing to recognize a three-syllable word from millions to billions of possible candidates. A novel interactive automatic-speech-recognition system is proposed to resolve this highly challenging task. This system was built as an open-vocabulary Chinese name recognition system using character-based approaches. Two important character-input speech-recognition modules were designed as backoff approaches in this system to complete the name input or to correct any misrecognized characters. Finite-state networks were compiled from regular grammar of syllable spellings and character descriptions for these two speech recognition modules. The possible candidate names cover more than five billions. This system has been tested publicly and proved a robust way to interact with the speaker. An 86.7% name recognition success rate was achieved by the interactive open-vocabulary Chinese name input system.

In this paper, we describe a parallel decoding-based ASR system developed of ATR that is robust to noise type, SNR and speaking style. It is difficult to recognize speech affected by various factors, especially when an ASR system contains only a single acoustic model. One solution is to employ multiple acoustic models, one model for each different condition. Even though the robustness of each acoustic model is limited, the whole ASR system can handle various conditions appropriately. In our system, there are two recognition sub-systems which use different features such as MFCC and Differential MFCC (DMFCC). Each sub-system has several acoustic models depending on SNR, speaker gender and speaking style, and during recognition each acoustic model is adapted by fast noise adaptation. From each sub-system, one hypothesis is selected based on posterior probability. The final recognition result is obtained by combining the best hypotheses from the two sub-systems. On the AURORA-2J task used widely for the evaluation of noise robustness, our system achieved higher recognition performance than a system which contains only a single model. Also, our system was tested using normal and hyper-articulated speech contaminated by several background noises, and exhibited high robustness to noise and speaking styles.

Minimum Bayes risk estimation and decoding strategies based on lattice segmentation techniques can be used to refine large vocabulary continuous speech recognition systems through the estimation of the parameters of the underlying hidden Markov models and through the identification of smaller recognition tasks which provides the opportunity to incorporate novel modeling and decoding procedures in LVCSR. These techniques are discussed in the context of going 'beyond HMMs', showing in particular that this process of subproblem identification makes it possible to train and apply small-domain binary pattern classifiers, such as Support Vector Machines, to large vocabulary continuous speech recognition.