Multilevel sampling and aggregation for discriminative training

End-to-end (E2E) models fold the acoustic, pronunciation and language models of a conventional speech recognition model into one neural network with a much smaller number of parameters than a conventional ASR system, thus making it suitable for on-device applications. For example, recurrent neural network transducer (RNN-T) as a streaming E2E model has shown promising potential for on-device ASR [1]. For such applications, quality and latency are two critical factors. We propose to reduce E2E model's latency by extending the RNN-T endpointer (RNN-T EP) model [2] with additional early and late penalties. By further applying the minimum word error rate (MWER) training technique [3], we achieved 8.0% relative word error rate (WER) reduction and 130ms 90-percentile latency reduction over [2] on a Voice Search test set. We also experimented with a second-pass Listen, Attend and Spell (LAS) rescorer [4]. Although it did not directly improve the first pass latency, the large WER reduction provides extra room to trade WER for latency. RNN-T EP+LAS, together with MWER training brings in 18.7% relative WER reduction and 160ms 90-percentile latency reductions compared to the original proposed RNN-T EP [2] model.

We propose a novel neural label embedding (NLE) scheme for the domain adaptation of a deep neural network (DNN) acoustic model with unpaired data samples from source and target domains. With NLE method, we distill the knowledge from a powerful source-domain DNN into a dictionary of label embeddings, or l-vectors, one for each senone class. Each l-vector is a representation of the senone-specific output distributions of the source-domain DNN and is learned to minimize the average L <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> , Kullback-Leibler (KL) or symmetric KL distance to the output vectors with the same label through simple averaging or standard back-propagation. During adaptation, the l-vectors serve as the soft targets to train the target-domain model with cross-entropy loss. Without parallel data constraint as in the teacher-student learning, NLE is specially suited for the situation where the paired target-domain data cannot be simulated from the source-domain data. We adapt a 6400 hours multi-conditional US English acoustic model to each of the 9 accented English (80 to 830 hours) and kids' speech (80 hours). NLE achieves up to 14.1% relative word error rate reduction over direct re-training with one-hot labels.

We present our study on semi-supervised Gaussian mixture model (GMM) hidden Markov model (HMM) and deep neural network (DNN) HMM acoustic model training. We analyze the impact of transcription quality and data sampling approach on the performance of the resulting model, and propose a multisystem combination and confidence re-calibration approach to improve the transcription inference and data selection. Compared to using a single system recognition result and confidence score, our proposed approach reduces the phone error rate of the inferred transcription by 23.8% relatively when top 60% of data are selected. Experiments were conducted on the mobile short message dictation (SMD) task. For the GMM-HMM model, we achieved 7.2% relative word error rate reduction (WERR) against a well-trained narrow-band fMPE+bMMI system by adding 2100 hours of untranscribed data, and 28.2% relative WERR over a wide-band MLE model trained from transcribed out-of-domain voice search data after adding 10K hours of untranscribed SMD data. For the CD-DNN-HMM model, 11.7% and 15.0% relative WERRs are achieved after adding 1K hours of untranscribed data using random and importance sampling, respectively. We also found using large amount of untranscribed data for pretraining does not help. Index Terms: semi-supervised acoustic model training, system combination, confidence re-calibration, importance sampling

We investigate several variants of speech-rate-dependent acoustic models for large-vocabulary conversational speech recognition, in the framework of combining rate-specific models in decoding to compensate for speech rate variation. We study two basic approaches to combining rate-specific models: one combines models at the pronunciation level and the other at the HMM state level. Furthermore, we investigate the influence of different numbers of rate-of-speech classes and different parameter tying schemes. Experiments on the Switchboard database, using SRI’s DECIPHER recognition system, show that rate-dependent acoustic modeling resulted in a 2% relative word error rate reduction over a rateindependent baseline, and that the pronunciation-level constraint, Gaussian sharing between rate-specific models, and a well-chosen number of rate-of-speech classes are all important for best performance.

In this paper we investigate the GMM-derived features for adaptation of context-dependent deep neural network HMM (CD-DNN-HMM) acoustic models with the focus on exploration of fusion of the adapted GMM-derived features and the conventional bottleneck features. We analyze and compare different types of fusion, such as feature level, posterior level, lattice level and others in order to discover the best possible way of fusion. Experimental results on the TED-LIUM corpus show that the proposed adaptation technique can be effectively integrated into DNN setup at different levels and provide additional gain in recognition performance: up to 6 % of relative word error rate reduction (WERR) over the strong speaker adapted DNN baseline, and up to 22 % of relative WERR in comparison with a speaker independent DNN baseline model, trained on conventional features.

In this paper, we present our initial efforts for building a code-switching (CS) speech recognition system leveraging existing acoustic models (AMs) and language models (LMs), i.e., no training required, and specifically targeting intra-sentential switching. To achieve such an ambitious goal, new mechanisms for foreign pronunciation generation and language model (LM) enrichment have been devised. Specifically, we have designed an automatic approach to obtain high quality pronunciation of foreign language (FL) words in the native language (NL) phoneme set using existing acoustic phone decoders and an LSTM-based grapheme-to-phoneme (G2P) model. Improved accented pronunciations have thus been obtained by learning foreign pronunciations directly from data. Furthermore, a code-switching LM was deployed by converting the original NL LM into a CS LM using translated word pairs and borrowing statistics for the NL LM. Experimental evidence clearly demonstrates that our approach better deals with accented foreign pronunciations than techniques based on human labeling. Moreover, our best system achieves a 55.5% relative word error rate reduction from 34.4%, obtained with a conventional monolingual ASR system, to 15.3% on an intra-sentential CS task without harming the monolingual recognition accuracy.

This paper proposes a simple yet effective unsupervised speaker adaptation approach for batch normalization based deep neural network acoustic models. The basic idea of this approach is to recompute means and variances in all batch normalization layers over the test data for every speaker. Thus the distribution of the test data can be close to the training data. This approach doesn't need to adjust any trainable parameters of the acoustic model. Experiments are conducted on CHiME-3 datasets. The results show that the proposed adaptation obtains improvement on the real test set by 2.17 % relative average word error rate (WER) reduction when compared with the scaling and shifting factors (SSF) adaptation. Combining our proposed MV adaptation with the SSF adaptation obtains further improvement.

Maximum a posteriori (MAP) adaptation and its discriminative variants, such as MMI-MAP (maximum mutual information MAP) and MPE-MAP (minimum phone error MAP), have been widely applied to acoustic model adaptation. This paper introduces a new adaptation approach, fMPE-MAP, which is an extension to the original fMPE (feature minimum phone error) algorithm, with the enhanced ability in porting Gaussian models and fMPE transforms to a new domain. We applied this approach to the SRI-ICSI 2007 NIST meeting recognition system, for which we ported our conversational telephone speech (CTS) and broadcast news (BN) models to the meeting domain. Experiments showed that the proposed fMPE-MAP approach has comparable or better performance than simply training the fMPE transform on combined data, in addition to the obvious speed advantage. In combination with MPE-MAP, we obtained about 20% relative word error rate reduction on a lecture meeting evaluation test set, over the models trained with the standard MAP approach.

A statistically designed Automatic Speech Recognition (ASR) system extracts features from speech signals using feature extraction methods, links the extracted features with the expected phonetics of the hypothesis using acoustic models, uses language model to add prior information about the structure of the target language. For many years, Mel-frequency Cepstral coefficients (MFCC), n-gram, and Hidden Markov Model (HMM) approaches have been used predominantly for feature extraction, language modeling and acoustic modeling, respectively. However, performance degradation of MFCC in noisy conditions and inaccuracy of HMMs while handling large vocabularies have made researchers to propose more efficient methods. The proposed work uses noise robust method Gammatone Frequency Cepstral Coefficients(GFCC) for feature extraction, trigram language modeling, and HMM-Gaussian mixture model (GMM) based acoustic modeling to implement a continuous Hindi language ASR system. Also, it applies Differential Evolution (DE) technique to refine the GFCC features and discriminative techniques to enhance performance of the acoustic model. The performance of the implemented system has been evaluated by using different feature extraction methods, variants of n-gram language modeling techniques and different discriminative techniques in clean as well as noisy conditions. Initially, the results reveal that DE optimized GFCC with HMM-Gaussian Mixture Model (GMM) acoustic modeling performs better than MFCC, PLP and MF-PLP feature extraction methods. Secondly, the experimental results show that the Minimum Phone Error (MPE) outperforms Maximum Mutual Information (MMI) and Maximum Likelihood Estimation (MLE) and trigram based language modeling gives more accurate results than unigram and bigram language modeling. Finally, it has been concluded that the continuous Hindi language ASR system implemented using DE refined GFFC feature extraction method with MPE discriminative training technique and trigram based language modeling gives better accuracy in clean as-well-as noisy environments.

Recent work on self-supervised pre-training focus on leveraging large-scale unlabeled speech data to build robust end-to-end (E2E) acoustic models (AM) that can be later fine-tuned on downstream tasks e.g., automatic speech recognition (ASR). Yet, few works investigated the impact on performance when the data properties substantially differ between the pre-training and fine-tuning phases, termed domain shift. We target this scenario by analyzing the robustness of Wav2Vec 2.0 and XLS-R models on downstream ASR for a completely unseen domain, air traffic control (ATC) communications. We benchmark these two models on several open-source and challenging ATC databases with signal-to-noise ratio between 5 to 20 dB. Relative word error rate (WER) reductions between 20% to 40% are obtained in comparison to hybrid-based ASR baselines by only fine-tuning E2E acoustic models with a smaller fraction of labeled data. We analyze WERs on the low-resource scenario and gender bias carried by one ATC dataset.

In this paper we investigate GMM-derived features recently introduced for adaptation of context-dependent deep neural network HMM (CD-DNN-HMM) acoustic models. We present an initial attempt of improving the previously proposed adaptation algorithm by applying lattice scores and by using confidence measures in the traditional maximum a posteriori (MAP) adaptation algorithm. Modified MAP adaptation is performed for the auxiliary GMM model used in a speaker adaptation procedure for a DNN. In addition we introduce two approaches - data augmentation and data selection, for improving the regularization in MAP adaptation for DNN. Experimental results on the Wall Street Journal (WSJ0) corpus show that the proposed adaptation technique can provide, on average, up to \(9.9\,\%\) relative word error rate (WER) reduction under an unsupervised adaptation setup, compared to speaker independent DNN-HMM systems built on conventional features.

In this work we compare two parameter optimization techniques for discriminative training using the MMI criterion: the extended Baum-Welch (EBW) algorithm and the generalized probabilistic descent (GPD) method. Using Gaussian emission densities we found special expressions for the step sizes in GPD, leading to reestimation formula very similar to those derived for the EBW algorithm. Results were produced for both the TI digitstring and the SieTill corpus for continuously spoken American English and German digitstrings. The results for both techniques do not show significant differences. This experimental results support the strong link between EBW and GPD as expected from the analytic comparison.

Machine Speech Chain integrates both end-to-end (E2E) automatic speech recognition (ASR) and neural text-to-speech (TTS) into one circle for joint training. It has been proven that it can effectively leverage a large amount of unpaired data in the spirit of data augmentation. In this paper, we explore the TTS→ASR pipeline in machine speech chain to perform domain adaptation for both E2E ASR and neural TTS models with only text data from the target domain. We conduct experiments by adapting from audiobook domain (i.e., LibriSpeech) to presentation domain (i.e., TED-LIUM). There is a relative word error rate (WER) reduction of 19.7% for the E2E ASR model on the TED-LIUM test set, and a relative WER reduction of 29.4% in synthetic speech generated by neural TTS in the presentation domain. Moreover, we observe that the gains from the proposed method and conventional adaptation methods of language models are additive.

Automatic speech recognition (ASR) in the cloud allows the use of larger models and more powerful multi-channel signal processing front-ends compared to on-device processing. However, it also adds an inherent latency due to the transmission of the audio signal, especially when transmitting multiple channels of a microphone array. One way to reduce the network bandwidth requirements is client-side compression with a lossy codec such as Opus. However, this compression can have a detrimental effect especially on multi-channel ASR front-ends, due to the distortion and loss of spatial information introduced by the codec. In this publication, we propose an improved approach for the compression of microphone array signals based on Opus, using a modified joint channel coding approach and additionally introducing a multi-channel spatial decorrelating transform to reduce redundancy in the transmission. We illustrate the effect of the proposed approach on the spatial information retained in multi-channel signals after compression, and evaluate the performance on far-field ASR with a multi-channel beamforming front-end. We demonstrate that our approach can lead to a 37.5 % bitrate reduction or a 5.1 % relative word error rate reduction for a fixed bitrate budget in a seven channel setup.

Long short-term memory (LSTM) recurrent neural networks (RNNs) have recently shown significant performance improvements over deep feed-forward neural networks (DNNs). A key aspect of these models is the use of time recurrence, combined with a gating architecture that ameliorates the vanishing gradient problem. Inspired by human spectrogram reading, in this paper we propose an extension to LSTMs that performs the recurrence in frequency as well as in time. This model first scans the frequency bands to generate a summary of the spectral information, and then uses the output layer activations as the input to a traditional time LSTM (T-LSTM). Evaluated on a Microsoft short message dictation task, the proposed model obtained a 3.6% relative word error rate reduction over the T-LSTM.