Automatic speech recognition systems have been largely improved in the past few decades and current systems are mainly hybrid-based and end-to-end-based. The recently proposed CTC-CRF framework inherits the data-efficiency of the hybrid approach and the simplicity of the end-to-end approach. In this paper, we further advance CTC-CRF based ASR technique with explorations on modeling units and neural architectures. Specifically, we investigate techniques to enable the recently developed wordpiece modeling units and Conformer neural networks to be succesfully applied in CTC-CRFs. Experiments are conducted on two English datasets (Switchboard, Librispeech) and a German dataset from CommonVoice. Experimental results suggest that (i) Conformer can improve the recognition performance significantly; (ii) Wordpiece-based systems perform slightly worse compared with phone-based systems for the target language with a low degree of grapheme-phoneme correspondence (e.g. English), while the two systems can perform equally strong when such degree of correspondence is high for the target language (e.g. German).
We employ a combination of recent developments in semi-supervised learning for automatic speech recognition to obtain state-of-the-art results on LibriSpeech utilizing the unlabeled audio of the Libri-Light dataset. More precisely, we carry out noisy student training with SpecAugment using giant Conformer models pre-trained using wav2vec 2.0 pre-training. By doing so, we are able to achieve word-error-rates (WERs) 1.4%/2.6% on the LibriSpeech test/test-other sets against the current state-of-the-art WERs 1.7%/3.3%.
The requirements for many applications of state-of-the-art speech recognition systems include not only low word error rate (WER) but also low latency. Specifically, for many use-cases, the system must be able to decode utterances in a streaming fashion and faster than real-time. Recently, a streaming recurrent neural network transducer (RNN-T) end-to-end (E2E) model has shown to be a good candidate for on-device speech recognition, with improved WER and latency metrics compared to conventional on-device models [1]. However, this model still lags behind a large state-of-the-art conventional model in quality [2]. On the other hand, a non-streaming E2E Listen, Attend and Spell (LAS) model has shown comparable quality to large conventional models [3]. This work aims to bring the quality of an E2E streaming model closer to that of a conventional system by incorporating a LAS network as a second-pass component, while still abiding by latency constraints. Our proposed two-pass model achieves a 17%-22% relative reduction in WER compared to RNN-T alone and increases latency by a small fraction over RNN-T.
In this paper, we propose a novel speech emotion recognition model called Cross Attention Network (CAN) that uses aligned audio and text signals as inputs. It is inspired by the fact that humans recognize speech as a combination of simultaneously produced acoustic and textual signals. First, our method segments the audio and the underlying text signals into equal number of steps in an aligned way so that the same time steps of the sequential signals cover the same time span in the signals. Together with this technique, we apply the cross attention to aggregate the sequential information from the aligned signals. In the cross attention, each modality is aggregated independently by applying the global attention mechanism onto each modality. Then, the attention weights of each modality are applied directly to the other modality in a crossed way, so that the CAN gathers the audio and text information from the same time steps based on each modality. In the experiments conducted on the standard IEMOCAP dataset, our model outperforms the state-of-the-art systems by 2.66% and 3.18% relatively in terms of the weighted and unweighted accuracy.
More recently, Bidirectional Encoder Representations from Transformers (BERT) was proposed and has achieved impressive success on many natural language processing (NLP) tasks such as question answering and language understanding, due mainly to its effective pre-training then fine-tuning paradigm as well as strong local contextual modeling ability. In view of the above, this paper presents a novel instantiation of the BERT-based contextualized language models (LMs) for use in reranking of N-best hypotheses produced by automatic speech recognition (ASR). To this end, we frame N-best hypothesis reranking with BERT as a prediction problem, which aims to predict the oracle hypothesis that has the lowest word error rate (WER) given the N-best hypotheses (denoted by PBERT). In particular, we also explore to capitalize on task-specific global topic information in an unsupervised manner to assist PBERT in N-best hypothesis reranking (denoted by TPBERT). Extensive experiments conducted on the AMI benchmark corpus demonstrate the effectiveness and feasibility of our methods in comparison to the conventional autoregressive models like the recurrent neural network (RNN) and a recently proposed method that employed BERT to compute pseudo-log-likelihood (PLL) scores for N-best hypothesis reranking.
Our study presents a series of experiments on speech recognition with endangered and extinct Samoyedic languages, spoken in Northern and Southern Siberia. To best of our knowledge, this is the first time a functional ASR system is built for an extinct language. We achieve with Kamas language a Label Error Rate of 15\%, and conclude through careful error analysis that this quality is already very useful as a starting point for refined human transcriptions. Our results with related Nganasan language are more modest, with best model having the error rate of 33\%. We show, however, through experiments where Kamas training data is enlarged incrementally, that Nganasan results are in line with what is expected under low-resource circumstances of the language. Based on this, we provide recommendations for scenarios in which further language documentation or archive processing activities could benefit from modern ASR technology. All training data and processing scripts haven been published on Zenodo with clear licences to ensure further work in this important topic.
The transducer architecture is becoming increasingly popular in the field of speech recognition, because it is naturally streaming as well as high in accuracy. One of the drawbacks of transducer is that it is difficult to decode in a fast and parallel way due to an unconstrained number of symbols that can be emitted per time step. In this work, we introduce a constrained version of transducer loss to learn strictly monotonic alignments between the sequences; we also improve the standard greedy search and beam search algorithms by limiting the number of symbols that can be emitted per time step in transducer decoding, making it more efficient to decode in parallel with batches. Furthermore, we propose an finite state automaton-based (FSA) parallel beam search algorithm that can run with graphs on GPU efficiently. The experiment results show that we achieve slight word error rate (WER) improvement as well as significant speedup in decoding. Our work is open-sourced and publicly available\footnote{https://github.com/k2-fsa/icefall}.
Despite significant efforts over the last few years to build a robust automatic speech recognition (ASR) system for different acoustic settings, the performance of the current state-of-the-art technologies significantly degrades in noisy reverberant environments. Convolutional Neural Networks (CNNs) have been successfully used to achieve substantial improvements in many speech processing applications including distant speech recognition (DSR). However, standard CNN architectures were not efficient in capturing long-term speech dynamics, which are essential in the design of a robust DSR system. In the present study, we address this issue by investigating variants of large receptive field CNNs (LRF-CNNs) which include deeply recursive networks, dilated convolutional neural networks, and stacked hourglass networks. To compare the efficacy of the aforementioned architectures with the standard CNN for Wall Street Journal (WSJ) corpus, we use a hybrid DNN-HMM based speech recognition system. We extend the study to evaluate the system performances for distant speech simulated using realistic room impulse responses (RIRs). Our experiments show that with fixed number of parameters across all architectures, the large receptive field networks show consistent improvements over the standard CNNs for distant speech. Amongst the explored LRF-CNNs, stacked hourglass network has shown improvements with a 8.9% relative reduction in word error rate (WER) and 10.7% relative improvement in frame accuracy compared to the standard CNNs for distant simulated speech signals.
Speech emotion recognition (SER) is an essential part of human-computer interaction. In this paper, we propose an SER network based on a Graph Isomorphism Network with Weighted Multiple Aggregators (WMA-GIN), which can effectively handle the problem of information confusion when neighbour nodes' features are aggregated together in GIN structure. Moreover, a Full-Adjacent (FA) layer is adopted for alleviating the over-squashing problem, which is existed in all Graph Neural Network (GNN) structures, including GIN. Furthermore, a multi-phase attention mechanism and multi-loss training strategy are employed to avoid missing the useful emotional information in the stacked WMA-GIN layers. We evaluated the performance of our proposed WMA-GIN on the popular IEMOCAP dataset. The experimental results show that WMA-GIN outperforms other GNN-based methods and is comparable to some advanced non-graph-based methods by achieving 72.48% of weighted accuracy (WA) and 67.72% of unweighted accuracy (UA).
Attention-based end-to-end automatic speech recognition (ASR) systems have recently demonstrated state-of-the-art results for numerous tasks. However, the application of self-attention and attention-based encoder-decoder models remains challenging for streaming ASR, where each word must be recognized shortly after it was spoken. In this work, we present the dual causal/non-causal self-attention (DCN) architecture, which in contrast to restricted self-attention prevents the overall context to grow beyond the look-ahead of a single layer when used in a deep architecture. DCN is compared to chunk-based and restricted self-attention using streaming transformer and conformer architectures, showing improved ASR performance over restricted self-attention and competitive ASR results compared to chunk-based self-attention, while providing the advantage of frame-synchronous processing. Combined with triggered attention, the proposed streaming end-to-end ASR systems obtained state-of-the-art results on the LibriSpeech, HKUST, and Switchboard ASR tasks.