Search

Speech-based automatic detection of Alzheimer's disease (AD) and depression has attracted increased attention. Confidence estimation is crucial for a trust-worthy automatic diagnostic system which informs the clinician about the confidence of model predictions and helps reduce the risk of misdiagnosis. This paper investigates confidence estimation for automatic detection of AD and depression based on clinical interviews. A novel Bayesian approach is proposed which uses a dynamic Dirichlet prior distribution to model the second-order probability of the predictive distribution. Experimental results on the publicly available ADReSS and DAIC-WOZ datasets demonstrate that the proposed method outperforms a range of baselines for both classification accuracy and confidence estimation., Comment: Accepted by Interspeech 2024

This paper introduces a novel approach to speaker-attributed ASR transcription using a neural clustering method. With a parallel processing mechanism, diarisation and ASR can be applied simultaneously, helping to prevent the accumulation of errors from one sub-system to the next in a cascaded system. This is achieved by the use of ASR, trained using a serialised output training method, together with segment-level discriminative neural clustering (SDNC) to assign speaker labels. With SDNC, our system does not require an extra non-neural clustering method to assign speaker labels, thus allowing the entire system to be based on neural networks. Experimental results on the AMI meeting dataset demonstrate that SDNC outperforms spectral clustering (SC) by a 19% relative diarisation error rate (DER) reduction on the AMI Eval set. When compared with the cascaded system with SC, the parallel system with SDNC gives a 7%/4% relative improvement in cpWER on the Dev/Eval set., Comment: To appear in Interspeech 2024

While the neural transducer is popular for online speech recognition, simultaneous speech translation (SST) requires both streaming and re-ordering capabilities. This paper presents the LS-Transducer-SST, a label-synchronous neural transducer for SST, which naturally possesses these two properties. The LS-Transducer-SST dynamically decides when to emit translation tokens based on an Auto-regressive Integrate-and-Fire (AIF) mechanism. A latency-controllable AIF is also proposed, which can control the quality-latency trade-off either only during decoding, or it can be used in both decoding and training. The LS-Transducer-SST can naturally utilise monolingual text-only data via its prediction network which helps alleviate the key issue of data sparsity for E2E SST. During decoding, a chunk-based incremental joint decoding technique is designed to refine and expand the search space. Experiments on the Fisher-CallHome Spanish (Es-En) and MuST-C En-De data show that the LS-Transducer-SST gives a better quality-latency trade-off than existing popular methods. For example, the LS-Transducer-SST gives a 3.1/2.9 point BLEU increase (Es-En/En-De) relative to CAAT at a similar latency and a 1.4 s reduction in average lagging latency with similar BLEU scores relative to Wait-k., Comment: Accepted by ACL 2024 Main Conference

Wav2Prompt is proposed which allows straightforward integration between spoken input and a text-based large language model (LLM). Wav2Prompt uses a simple training process with only the same data used to train an automatic speech recognition (ASR) model. After training, Wav2Prompt learns continuous representations from speech and uses them as LLM prompts. To avoid task over-fitting issues found in prior work and preserve the emergent abilities of LLMs, Wav2Prompt takes LLM token embeddings as the training targets and utilises a continuous integrate-and-fire mechanism for explicit speech-text alignment. Therefore, a Wav2Prompt-LLM combination can be applied to zero-shot spoken language tasks such as speech translation (ST), speech understanding (SLU), speech question answering (SQA) and spoken-query-based QA (SQQA). It is shown that for these zero-shot tasks, Wav2Prompt performs similarly to an ASR-LLM cascade and better than recent prior work. If relatively small amounts of task-specific paired data are available in few-shot scenarios, the Wav2Prompt-LLM combination can be end-to-end (E2E) fine-tuned. The Wav2Prompt-LLM combination then yields greatly improved results relative to an ASR-LLM cascade for the above tasks. For instance, for English-French ST with the BLOOMZ-7B1 LLM, a Wav2Prompt-LLM combination gave a 8.5 BLEU point increase over an ASR-LLM cascade.

Speech emotion recognition is a challenging classification task with natural emotional speech, especially when the distribution of emotion types is imbalanced in the training and test data. In this case, it is more difficult for a model to learn to separate minority classes, resulting in those sometimes being ignored or frequently misclassified. Previous work has utilised class weighted loss for training, but problems remain as it sometimes causes over-fitting for minor classes or under-fitting for major classes. This paper presents the system developed by a multi-site team for the participation in the Odyssey 2024 Emotion Recognition Challenge Track-1. The challenge data has the aforementioned properties and therefore the presented systems aimed to tackle these issues, by introducing focal loss in optimisation when applying class weighted loss. Specifically, the focal loss is further weighted by prior-based class weights. Experimental results show that combining these two approaches brings better overall performance, by sacrificing performance on major classes. The system further employs a majority voting strategy to combine the outputs of an ensemble of 7 models. The models are trained independently, using different acoustic features and loss functions - with the aim to have different properties for different data. Hence these models show different performance preferences on major classes and minor classes. The ensemble system output obtained the best performance in the challenge, ranking top-1 among 68 submissions. It also outperformed all single models in our set. On the Odyssey 2024 Emotion Recognition Challenge Task-1 data the system obtained a Macro-F1 score of 35.69% and an accuracy of 37.32%.

The subjective perception of emotion leads to inconsistent labels from human annotators. Typically, utterances lacking majority-agreed labels are excluded when training an emotion classifier, which cause problems when encountering ambiguous emotional expressions during testing. This paper investigates three methods to handle ambiguous emotion. First, we show that incorporating utterances without majority-agreed labels as an additional class in the classifier reduces the classification performance of the other emotion classes. Then, we propose detecting utterances with ambiguous emotions as out-of-domain samples by quantifying the uncertainty in emotion classification using evidential deep learning. This approach retains the classification accuracy while effectively detects ambiguous emotion expressions. Furthermore, to obtain fine-grained distinctions among ambiguous emotions, we propose representing emotion as a distribution instead of a single class label. The task is thus re-framed from classification to distribution estimation where every individual annotation is taken into account, not just the majority opinion. The evidential uncertainty measure is extended to quantify the uncertainty in emotion distribution estimation. Experimental results on the IEMOCAP and CREMA-D datasets demonstrate the superior capability of the proposed method in terms of majority class prediction, emotion distribution estimation, and uncertainty estimation.

Foundation models have shown superior performance for speech emotion recognition (SER). However, given the limited data in emotion corpora, finetuning all parameters of large pre-trained models for SER can be both resource-intensive and susceptible to overfitting. This paper investigates parameter-efficient finetuning (PEFT) for SER. Various PEFT adaptors are systematically studied for both classification of discrete emotion categories and prediction of dimensional emotional attributes. The results demonstrate that the combination of PEFT methods surpasses full finetuning with a significant reduction in the number of trainable parameters. Furthermore, a two-stage adaptation strategy is proposed to adapt models trained on acted emotion data, which is more readily available, to make the model more adept at capturing natural emotional expressions. Both intra- and cross-corpus experiments validate the efficacy of the proposed approach in enhancing the performance on both the source and target domains.

Recently, connectionist temporal classification (CTC)-based end-to-end (E2E) automatic speech recognition (ASR) models have achieved impressive results, especially with the development of self-supervised learning. However, E2E ASR models trained on paired speech-text data often suffer from domain shifts from training to testing. To alleviate this issue, this paper proposes a flat-start joint training method, named FastInject, which efficiently injects multi-domain unpaired text data into CTC-based ASR training. To maintain training efficiency, text units are pre-upsampled, and their representations are fed into the CTC model along with speech features. To bridge the modality gap between speech and text, an attention-based modality matching mechanism (AM3) is proposed, which retains the E2E flat-start training. Experiments show that the proposed FastInject gave a 22\% relative WER reduction (WERR) for intra-domain Librispeech-100h data and 20\% relative WERR on out-of-domain test sets., Comment: Accepted by ICASSP2024

Although end-to-end (E2E) automatic speech recognition (ASR) has shown state-of-the-art recognition accuracy, it tends to be implicitly biased towards the training data distribution which can degrade generalisation. This paper proposes a label-synchronous neural transducer (LS-Transducer), which provides a natural approach to domain adaptation based on text-only data. The LS-Transducer extracts a label-level encoder representation before combining it with the prediction network output. Since blank tokens are no longer needed, the prediction network performs as a standard language model, which can be easily adapted using text-only data. An Auto-regressive Integrate-and-Fire (AIF) mechanism is proposed to generate the label-level encoder representation while retaining low latency operation that can be used for streaming. In addition, a streaming joint decoding method is designed to improve ASR accuracy while retaining synchronisation with AIF. Experiments show that compared to standard neural transducers, the proposed LS-Transducer gave a 12.9% relative WER reduction (WERR) for intra-domain LibriSpeech data, as well as 21.4% and 24.6% relative WERRs on cross-domain TED-LIUM 2 and AESRC2020 data with an adapted prediction network., Comment: This work has been submitted to the IEEE for possible publication

Recently, advancements in large language models (LLMs) have shown an unprecedented ability across various language tasks. This paper investigates the potential application of LLMs to slot filling with noisy ASR transcriptions, via both in-context learning and task-specific fine-tuning. Dedicated prompt designs and fine-tuning approaches are proposed to improve the robustness of LLMs for slot filling with noisy ASR transcriptions. Moreover, a linearised knowledge injection (LKI) scheme is also proposed to integrate dynamic external knowledge into LLMs. Experiments were performed on SLURP to quantify the performance of LLMs, including GPT-3.5-turbo, GPT-4, LLaMA-13B and Vicuna-13B (v1.1 and v1.5) with different ASR error rates. The use of the proposed fine-tuning together with the LKI scheme for LLaMA-13B achieved an 8.3% absolute SLU-F1 improvement compared to the strong Flan-T5-base baseline system on a limited data setup.

We propose DiffSpEx, a generative target speaker extraction method based on score-based generative modelling through stochastic differential equations. DiffSpEx deploys a continuous-time stochastic diffusion process in the complex short-time Fourier transform domain, starting from the target speaker source and converging to a Gaussian distribution centred on the mixture of sources. For the reverse-time process, a parametrised score function is conditioned on a target speaker embedding to extract the target speaker from the mixture of sources. We utilise ECAPA-TDNN target speaker embeddings and condition the score function alternately on the SDE time embedding and the target speaker embedding. The potential of DiffSpEx is demonstrated with the WSJ0-2mix dataset, achieving an SI-SDR of 12.9 dB and a NISQA score of 3.56. Moreover, we show that fine-tuning a pre-trained DiffSpEx model to a specific speaker further improves performance, enabling personalisation in target speaker extraction., Comment: 5 pages, 4 figures, submitted to ICASSP 2024

Human annotator simulation (HAS) serves as a cost-effective substitute for human evaluation such as data annotation and system assessment. Human perception and behaviour during human evaluation exhibit inherent variability due to diverse cognitive processes and subjective interpretations, which should be taken into account in modelling to better mimic the way people perceive and interact with the world. This paper introduces a novel meta-learning framework that treats HAS as a zero-shot density estimation problem, which incorporates human variability and allows for the efficient generation of human-like annotations for unlabelled test inputs. Under this framework, we propose two new model classes, conditional integer flows and conditional softmax flows, to account for ordinal and categorical annotations, respectively. The proposed method is evaluated on three real-world human evaluation tasks and shows superior capability and efficiency to predict the aggregated behaviours of human annotators, match the distribution of human annotations, and simulate the inter-annotator disagreements., Comment: Code available at: https://github.com/W-Wu/HAS_CNF

Although end-to-end (E2E) trainable automatic speech recognition (ASR) has shown great success by jointly learning acoustic and linguistic information, it still suffers from the effect of domain shifts, thus limiting potential applications. The E2E ASR model implicitly learns an internal language model (LM) which characterises the training distribution of the source domain, and the E2E trainable nature makes the internal LM difficult to adapt to the target domain with text-only data To solve this problem, this paper proposes decoupled structures for attention-based encoder-decoder (Decoupled-AED) and neural transducer (Decoupled-Transducer) models, which can achieve flexible domain adaptation in both offline and online scenarios while maintaining robust intra-domain performance. To this end, the acoustic and linguistic parts of the E2E model decoder (or prediction network) are decoupled, making the linguistic component (i.e. internal LM) replaceable. When encountering a domain shift, the internal LM can be directly replaced during inference by a target-domain LM, without re-training or using domain-specific paired speech-text data. Experiments for E2E ASR models trained on the LibriSpeech-100h corpus showed that the proposed decoupled structure gave 15.1% and 17.2% relative word error rate reductions on the TED-LIUM 2 and AESRC2020 corpora while still maintaining performance on intra-domain data.

Although automatic emotion recognition (AER) has recently drawn significant research interest, most current AER studies use manually segmented utterances, which are usually unavailable for dialogue systems. This paper proposes integrating AER with automatic speech recognition (ASR) and speaker diarisation (SD) in a jointly-trained system. Distinct output layers are built for four sub-tasks including AER, ASR, voice activity detection and speaker classification based on a shared encoder. Taking the audio of a conversation as input, the integrated system finds all speech segments and transcribes the corresponding emotion classes, word sequences, and speaker identities. Two metrics are proposed to evaluate AER performance with automatic segmentation based on time-weighted emotion and speaker classification errors. Results on the IEMOCAP dataset show that the proposed system consistently outperforms two baselines with separately trained single-task systems on AER, ASR and SD., Comment: Interspeech 2023

Neural transducers provide a natural way of streaming ASR. However, they augment output sequences with blank tokens which leads to challenges for domain adaptation using text data. This paper proposes a label-synchronous neural transducer (LS-Transducer), which extracts a label-level encoder representation before combining it with the prediction network output. Hence blank tokens are no longer needed and the prediction network can be easily adapted using text data. An Auto-regressive Integrate-and-Fire (AIF) mechanism is proposed to generate the label-level encoder representation while retaining the streaming property. In addition, a streaming joint decoding method is designed to improve ASR accuracy. Experiments show that compared to standard neural transducers, the proposed LS-Transducer gave a 10% relative WER reduction (WERR) for intra-domain Librispeech-100h data, as well as 17% and 19% relative WERRs on cross-domain TED-LIUM2 and AESRC2020 data with an adapted prediction network.

Manually annotating fine-grained slot-value labels for task-oriented dialogue (ToD) systems is an expensive and time-consuming endeavour. This motivates research into slot-filling methods that operate with limited amounts of labelled data. Moreover, the majority of current work on ToD is based solely on text as the input modality, neglecting the additional challenges of imperfect automatic speech recognition (ASR) when working with spoken language. In this work, we propose a Knowledge-Aware Audio-Grounded generative slot-filling framework, termed KA2G, that focuses on few-shot and zero-shot slot filling for ToD with speech input. KA2G achieves robust and data-efficient slot filling for speech-based ToD by 1) framing it as a text generation task, 2) grounding text generation additionally in the audio modality, and 3) conditioning on available external knowledge (e.g. a predefined list of possible slot values). We show that combining both modalities within the KA2G framework improves the robustness against ASR errors. Further, the knowledge-aware slot-value generator in KA2G, implemented via a pointer generator mechanism, particularly benefits few-shot and zero-shot learning. Experiments, conducted on the standard speech-based single-turn SLURP dataset and a multi-turn dataset extracted from a commercial ToD system, display strong and consistent gains over prior work, especially in few-shot and zero-shot setups., Comment: to submit to CS&L

In automatic emotion recognition (AER), labels assigned by different human annotators to the same utterance are often inconsistent due to the inherent complexity of emotion and the subjectivity of perception. Though deterministic labels generated by averaging or voting are often used as the ground truth, it ignores the intrinsic uncertainty revealed by the inconsistent labels. This paper proposes a Bayesian approach, deep evidential emotion regression (DEER), to estimate the uncertainty in emotion attributes. Treating the emotion attribute labels of an utterance as samples drawn from an unknown Gaussian distribution, DEER places an utterance-specific normal-inverse gamma prior over the Gaussian likelihood and predicts its hyper-parameters using a deep neural network model. It enables a joint estimation of emotion attributes along with the aleatoric and epistemic uncertainties. AER experiments on the widely used MSP-Podcast and IEMOCAP datasets showed DEER produced state-of-the-art results for both the mean values and the distribution of emotion attributes., Comment: Accepted by ACL 2023

End-to-end automatic speech recognition (ASR) and large language models, such as Whisper and GPT-2, have recently been scaled to use vast amounts of training data. Despite the large amount of training data, infrequent content words that occur in a particular task may still exhibit poor ASR performance, with contextual biasing a possible remedy. This paper investigates the effectiveness of neural contextual biasing for Whisper combined with GPT-2. Specifically, this paper proposes integrating an adapted tree-constrained pointer generator (TCPGen) component for Whisper and a dedicated training scheme to dynamically adjust the final output without modifying any Whisper model parameters. Experiments across three datasets show a considerable reduction in errors on biasing words with a biasing list of 1000 words. Contextual biasing was more effective when applied to domain-specific data and can boost the performance of Whisper and GPT-2 without losing their generality., Comment: To appear in Interspeech 2023

This paper proposes handling training data sparsity in speech-based automatic depression detection (SDD) using foundation models pre-trained with self-supervised learning (SSL). An analysis of SSL representations derived from different layers of pre-trained foundation models is first presented for SDD, which provides insight to suitable indicator for depression detection. Knowledge transfer is then performed from automatic speech recognition (ASR) and emotion recognition to SDD by fine-tuning the foundation models. Results show that the uses of oracle and ASR transcriptions yield similar SDD performance when the hidden representations of the ASR model is incorporated along with the ASR textual information. By integrating representations from multiple foundation models, state-of-the-art SDD results based on real ASR were achieved on the DAIC-WOZ dataset.

Source separation can improve automatic speech recognition (ASR) under multi-party meeting scenarios by extracting single-speaker signals from overlapped speech. Despite the success of self-supervised learning models in single-channel source separation, most studies have focused on simulated setups. In this paper, seven SSL models were compared on both simulated and real-world corpora. Then, we propose to integrate the best-performing model WavLM into an automatic transcription system through a novel iterative source selection method. To improve real-world performance, time-domain unsupervised mixture invariant training was adapted to the time-frequency domain. Experiments showed that in the transcription system when source separation was inserted before an ASR model fine-tuned on separated speech, absolute reductions of 1.9% and 1.5% in concatenated minimum-permutation word error rate for an unknown number of speakers (cpWER-us) were observed on the AMI dev and test sets., Comment: To appear in Proc. ICASSP2023

Although large foundation models pre-trained by self-supervised learning have achieved state-of-the-art performance in many tasks including automatic speech recognition (ASR), knowledge distillation (KD) is often required in practice to transfer the knowledge learned by large teacher models into much smaller student models with affordable computation and memory costs. This paper proposes a novel two-stage KD framework to distil the knowledge from multiple speech foundation models as teachers into a single student neural transducer model for ASR. In the first stage, the student model encoder is pre-trained using the embeddings extracted from multiple teacher models. In the second stage, the student encoder is fine-tuned with the audio-text pairs based on the ASR task. Experiments on the LibriSpeech 100-hour subset show that the proposed KD framework improves the performance of both streaming and non-streaming student models when using only one teacher. The performance of the student model can be further enhanced when multiple teachers are used jointly, achieving word error rate reductions (WERRs) of 17.5% and 10.6%. Our proposed framework can be combined with other existing KD methods to achieve further improvements. Further WERRs were obtained by incorporating extra unlabelled data during encoder pre-training, leading to a total relative WERR of 55.0% on the non-streaming student model.

End-to-end (E2E) automatic speech recognition (ASR) implicitly learns the token sequence distribution of paired audio-transcript training data. However, it still suffers from domain shifts from training to testing, and domain adaptation is still challenging. To alleviate this problem, this paper designs a replaceable internal language model (RILM) method, which makes it feasible to directly replace the internal language model (LM) of E2E ASR models with a target-domain LM in the decoding stage when a domain shift is encountered. Furthermore, this paper proposes a residual softmax (R-softmax) that is designed for CTC-based E2E ASR models to adapt to the target domain without re-training during inference. For E2E ASR models trained on the LibriSpeech corpus, experiments showed that the proposed methods gave a 2.6% absolute WER reduction on the Switchboard data and a 1.0% WER reduction on the AESRC2020 corpus while maintaining intra-domain ASR results., Comment: Accepted by ICASSP2023

Automatic emotion recognition in conversation (ERC) is crucial for emotion-aware conversational artificial intelligence. This paper proposes a distribution-based framework that formulates ERC as a sequence-to-sequence problem for emotion distribution estimation. The inherent ambiguity of emotions and the subjectivity of human perception lead to disagreements in emotion labels, which is handled naturally in our framework from the perspective of uncertainty estimation in emotion distributions. A Bayesian training loss is introduced to improve the uncertainty estimation by conditioning each emotional state on an utterance-specific Dirichlet prior distribution. Experimental results on the IEMOCAP dataset show that ERC outperformed the single-utterance-based system, and the proposed distribution-based ERC methods have not only better classification accuracy, but also show improved uncertainty estimation., Comment: To appear in SLT 2022

Self-supervised learning via masked prediction pre-training (MPPT) has shown impressive performance on a range of speech-processing tasks. This paper proposes a method to bias self-supervised learning towards a specific task. The core idea is to slightly finetune the model that is used to obtain the target sequence. This leads to better performance and a substantial increase in training speed. Furthermore, this paper proposes a variant of MPPT that allows low-footprint streaming models to be trained effectively by computing the MPPT loss on masked and unmasked frames. These approaches are evaluated for automatic speech recognition on the Librispeech corpus, where 100 hours of data served as the labelled data and 860 hours as the unlabelled data. The biased training outperforms the unbiased training by 15.5% after 250k updates and 23.8% after 100k updates on test-other. For the streaming models, the pre-training approach yields a reduction in word error rate of 44.1%., Comment: Submitted to ICASSP 2023

End-to-end spoken language understanding (SLU) suffers from the long-tail word problem. This paper exploits contextual biasing, a technique to improve the speech recognition of rare words, in end-to-end SLU systems. Specifically, a tree-constrained pointer generator (TCPGen), a powerful and efficient biasing model component, is studied, which leverages a slot shortlist with corresponding entities to extract biasing lists. Meanwhile, to bias the SLU model output slot distribution, a slot probability biasing (SPB) mechanism is proposed to calculate a slot distribution from TCPGen. Experiments on the SLURP dataset showed consistent SLU-F1 improvements using TCPGen and SPB, especially on unseen entities. On a new split by holding out 5 slot types for the test, TCPGen with SPB achieved zero-shot learning with an SLU-F1 score over 50% compared to baselines which can not deal with it. In addition to slot filling, the intent classification accuracy was also improved., Comment: 5 pages, to appear in ICASSP 2023

In speaker diarisation, speaker embedding extraction models often suffer from the mismatch between their training loss functions and the speaker clustering method. In this paper, we propose the method of spectral clustering-aware learning of embeddings (SCALE) to address the mismatch. Specifically, besides an angular prototype cal (AP) loss, SCALE uses a novel affinity matrix loss which directly minimises the error between the affinity matrix estimated from speaker embeddings and the reference. SCALE also includes p-percentile thresholding and Gaussian blur as two important hyper-parameters for spectral clustering in training. Experiments on the AMI dataset showed that speaker embeddings obtained with SCALE achieved over 50% relative speaker error rate reductions using oracle segmentation, and over 30% relative diarisation error rate reductions using automatic segmentation when compared to a strong baseline with the AP-loss-based speaker embeddings., Comment: To appear in ICASSP 2023, 5 pages

Self-supervised-learning-based pre-trained models for speech data, such as Wav2Vec 2.0 (W2V2), have become the backbone of many speech tasks. In this paper, to achieve speaker diarisation and speech recognition using a single model, a tandem multitask training (TMT) method is proposed to fine-tune W2V2. For speaker diarisation, the tasks of voice activity detection (VAD) and speaker classification (SC) are required, and connectionist temporal classification (CTC) is used for ASR. The multitask framework implements VAD, SC, and ASR using an early layer, middle layer, and late layer of W2V2, which coincides with the order of segmenting the audio with VAD, clustering the segments based on speaker embeddings, and transcribing each segment with ASR. Experimental results on the augmented multi-party (AMI) dataset showed that using different W2V2 layers for VAD, SC, and ASR from the earlier to later layers for TMT not only saves computational cost, but also reduces diarisation error rates (DERs). Joint fine-tuning of VAD, SC, and ASR yielded 16%/17% relative reductions of DER with manual/automatic segmentation respectively, and consistent reductions in speaker attributed word error rate, compared to the baseline with separately fine-tuned models., Comment: To appear in Interspeech 2022

Incorporating biasing words obtained as contextual knowledge is critical for many automatic speech recognition (ASR) applications. This paper proposes the use of graph neural network (GNN) encodings in a tree-constrained pointer generator (TCPGen) component for end-to-end contextual ASR. By encoding the biasing words in the prefix-tree with a tree-based GNN, lookahead for future wordpieces in end-to-end ASR decoding is achieved at each tree node by incorporating information about all wordpieces on the tree branches rooted from it, which allows a more accurate prediction of the generation probability of the biasing words. Systems were evaluated on the Librispeech corpus using simulated biasing tasks, and on the AMI corpus by proposing a novel visual-grounded contextual ASR pipeline that extracts biasing words from slides alongside each meeting. Results showed that TCPGen with GNN encodings achieved about a further 15% relative WER reduction on the biasing words compared to the original TCPGen, with a negligible increase in the computation cost for decoding., Comment: To appear in Interspeech 2022. arXiv admin note: text overlap with arXiv:2205.09058

Contextual knowledge is essential for reducing speech recognition errors on high-valued long-tail words. This paper proposes a novel tree-constrained pointer generator (TCPGen) component that enables end-to-end ASR models to bias towards a list of long-tail words obtained using external contextual information. With only a small overhead in memory use and computation cost, TCPGen can structure thousands of biasing words efficiently into a symbolic prefix-tree and creates a neural shortcut between the tree and the final ASR output to facilitate the recognition of the biasing words. To enhance TCPGen, we further propose a novel minimum biasing word error (MBWE) loss that directly optimises biasing word errors during training, along with a biasing-word-driven language model discounting (BLMD) method during the test. All contextual ASR systems were evaluated on the public Librispeech audiobook corpus and the data from the dialogue state tracking challenges (DSTC) with the biasing lists extracted from the dialogue-system ontology. Consistent word error rate (WER) reductions were achieved with TCPGen, which were particularly significant on the biasing words with around 40\% relative reductions in the recognition error rates. MBWE and BLMD further improved the effectiveness of TCPGen and achieved more significant WER reductions on the biasing words. TCPGen also achieved zero-shot learning of words not in the audio training set with large WER reductions on the out-of-vocabulary words in the biasing list., Comment: This work has been submitted to the IEEE Transactions on Audio, Speech, and Language Processing for possible publication

Emotion recognition is a key attribute for artificial intelligence systems that need to naturally interact with humans. However, the task definition is still an open problem due to the inherent ambiguity of emotions. In this paper, a novel Bayesian training loss based on per-utterance Dirichlet prior distributions is proposed for verbal emotion recognition, which models the uncertainty in one-hot labels created when human annotators assign the same utterance to different emotion classes. An additional metric is used to evaluate the performance by detection test utterances with high labelling uncertainty. This removes a major limitation that emotion classification systems only consider utterances with labels where the majority of annotators agree on the emotion class. Furthermore, a frequentist approach is studied to leverage the continuous-valued "soft" labels obtained by averaging the one-hot labels. We propose a two-branch model structure for emotion classification on a per-utterance basis, which achieves state-of-the-art classification results on the widely used IEMOCAP dataset. Based on this, uncertainty estimation experiments were performed. The best performance in terms of the area under the precision-recall curve when detecting utterances with high uncertainty was achieved by interpolating the Bayesian training loss with the Kullback-Leibler divergence training loss for the soft labels. The generality of the proposed approach was verified using the MSP-Podcast dataset which yielded the same pattern of results.

Self-supervised pre-training is an effective approach to leveraging a large amount of unlabelled data to reduce word error rates (WERs) of automatic speech recognition (ASR) systems. Since it is impractical to use large pre-trained models for many real-world ASR applications, it is desirable to have a much smaller model while retaining the performance of the pre-trained model. In this paper, we propose a simple knowledge distillation (KD) loss function for neural transducers that focuses on the one-best path in the output probability lattice under both streaming and non-streaming setups, which allows a small student model to approach the performance of the large pre-trained teacher model. Experiments on the LibriSpeech dataset show that despite being 10 times smaller than the teacher model, the proposed loss results in relative WER reductions (WERRs) of 11.5% and 6.8% on the test-other set for non-streaming and streaming student models compared to the baseline transducers trained without KD using the labelled 100-hour clean data. With an additional 860 hours of unlabelled data for KD, the WERRs increase to 48.2% and 38.5% for non-streaming and streaming students. If language model shallow fusion is used for producing distillation targets, a further improvement in the student model is observed., Comment: Accepted as a conference paper at ICASSP 2022

As end-to-end automatic speech recognition (ASR) models reach promising performance, various downstream tasks rely on good confidence estimators for these systems. Recent research has shown that model-based confidence estimators have a significant advantage over using the output softmax probabilities. If the input data to the speech recogniser is from mismatched acoustic and linguistic conditions, the ASR performance and the corresponding confidence estimators may exhibit severe degradation. Since confidence models are often trained on the same in-domain data as the ASR, generalising to out-of-domain (OOD) scenarios is challenging. By keeping the ASR model untouched, this paper proposes two approaches to improve the model-based confidence estimators on OOD data: using pseudo transcriptions and an additional OOD language model. With an ASR model trained on LibriSpeech, experiments show that the proposed methods can greatly improve the confidence metrics on TED-LIUM and Switchboard datasets while preserving in-domain performance. Furthermore, the improved confidence estimators are better calibrated on OOD data and can provide a much more reliable criterion for data selection., Comment: Accepted as a conference paper at ICASSP 2022

Contextual knowledge is important for real-world automatic speech recognition (ASR) applications. In this paper, a novel tree-constrained pointer generator (TCPGen) component is proposed that incorporates such knowledge as a list of biasing words into both attention-based encoder-decoder and transducer end-to-end ASR models in a neural-symbolic way. TCPGen structures the biasing words into an efficient prefix tree to serve as its symbolic input and creates a neural shortcut between the tree and the final ASR output distribution to facilitate recognising biasing words during decoding. Systems were trained and evaluated on the Librispeech corpus where biasing words were extracted at the scales of an utterance, a chapter, or a book to simulate different application scenarios. Experimental results showed that TCPGen consistently improved word error rates (WERs) compared to the baselines, and in particular, achieved significant WER reductions on the biasing words. TCPGen is highly efficient: it can handle 5,000 biasing words and distractors and only add a small overhead to memory use and computation cost., Comment: To appear in ASRU 2021

Language models (LMs) pre-trained on massive amounts of text, in particular bidirectional encoder representations from Transformers (BERT), generative pre-training (GPT), and GPT-2, have become a key technology for many natural language processing tasks. In this paper, we present results using fine-tuned GPT, GPT-2, and their combination for automatic speech recognition (ASR). Unlike unidirectional LM GPT and GPT-2, BERT is bidirectional whose direct product of the output probabilities is no longer a valid language prior probability. A conversion method is proposed to compute the correct language prior probability based on bidirectional LM outputs in a mathematically exact way. Experimental results on the widely used AMI and Switchboard ASR tasks showed that the combination of the fine-tuned GPT and GPT-2 outperformed the combination of three neural LMs with different architectures trained from scratch on the in-domain text by up to a 12% relative word error rate reduction (WERR). Furthermore, on the AMI corpus, the proposed conversion for language prior probabilities enables BERT to obtain an extra 3% relative WERR, and the combination of BERT, GPT and GPT-2 results in further improvements., Comment: To appear in ASRU 2021

Commonly used automatic speech recognition (ASR) systems can be classified into frame-synchronous and label-synchronous categories, based on whether the speech is decoded on a per-frame or per-label basis. Frame-synchronous systems, such as traditional hidden Markov model systems, can easily incorporate existing knowledge and can support streaming ASR applications. Label-synchronous systems, based on attention-based encoder-decoder models, can jointly learn the acoustic and language information with a single model, which can be regarded as audio-grounded language models. In this paper, we propose rescoring the N-best hypotheses or lattices produced by a first-pass frame-synchronous system with a label-synchronous system in a second-pass. By exploiting the complementary modelling of the different approaches, the combined two-pass systems achieve competitive performance without using any extra speech or text data on two standard ASR tasks. For the 80-hour AMI IHM dataset, the combined system has a 13.7% word error rate (WER) on the evaluation set, which is up to a 29% relative WER reduction over the individual systems. For the 300-hour Switchboard dataset, the WERs of the combined system are 5.7% and 12.1% on Switchboard and CallHome subsets of Hub5'00, and 13.2% and 7.6% on Switchboard Cellular and Fisher subsets of RT03, up to a 33% relative reduction in WER over the individual systems., Comment: Submitted to IEEE/ACM Transactions on Audio Speech and Language Processing

End-to-end models with auto-regressive decoders have shown impressive results for automatic speech recognition (ASR). These models formulate the sequence-level probability as a product of the conditional probabilities of all individual tokens given their histories. However, the performance of locally normalised models can be sub-optimal because of factors such as exposure bias. Consequently, the model distribution differs from the underlying data distribution. In this paper, the residual energy-based model (R-EBM) is proposed to complement the auto-regressive ASR model to close the gap between the two distributions. Meanwhile, R-EBMs can also be regarded as utterance-level confidence estimators, which may benefit many downstream tasks. Experiments on a 100hr LibriSpeech dataset show that R-EBMs can reduce the word error rates (WERs) by 8.2%/6.7% while improving areas under precision-recall curves of confidence scores by 12.6%/28.4% on test-clean/test-other sets. Furthermore, on a state-of-the-art model using self-supervised learning (wav2vec 2.0), R-EBMs still significantly improves both the WER and confidence estimation performance., Comment: To appear in Proc. Interspeech 2021

This paper presents a novel natural gradient and Hessian-free (NGHF) optimisation framework for neural network training that can operate efficiently in a distributed manner. It relies on the linear conjugate gradient (CG) algorithm to combine the natural gradient (NG) method with local curvature information from Hessian-free (HF) or other second-order methods. A solution to a numerical issue in CG allows effective parameter updates to be generated with far fewer CG iterations than usually used (e.g. 5-8 instead of 200). This work also presents a novel preconditioning approach to improve the progress made by individual CG iterations for models with shared parameters. Although applicable to other training losses and model structures, NGHF is investigated in this paper for lattice-based discriminative sequence training for hybrid hidden Markov model acoustic models using a standard recurrent neural network, long short-term memory, and time delay neural network models for output probability calculation. Automatic speech recognition experiments are reported on the multi-genre broadcast data set for a range of different acoustic model types. These experiments show that NGHF achieves larger word error rate reductions than standard stochastic gradient descent or Adam, while requiring orders of magnitude fewer parameter updates.

The effective incorporation of cross-utterance information has the potential to improve language models (LMs) for automatic speech recognition (ASR). To extract more powerful and robust cross-utterance representations for the Transformer LM (TLM), this paper proposes the R-TLM which uses hidden states in a long short-term memory (LSTM) LM. To encode the cross-utterance information, the R-TLM incorporates an LSTM module together with a segment-wise recurrence in some of the Transformer blocks. In addition to the LSTM module output, a shortcut connection using a fusion layer that bypasses the LSTM module is also investigated. The proposed system was evaluated on the AMI meeting corpus, the Eval2000 and the RT03 telephone conversation evaluation sets. The best R-TLM achieved 0.9%, 0.6%, and 0.8% absolute WER reductions over the single-utterance TLM baseline, and 0.5%, 0.3%, 0.2% absolute WER reductions over a strong cross-utterance TLM baseline on the AMI evaluation set, Eval2000 and RT03 respectively. Improvements on Eval2000 and RT03 were further supported by significance tests. R-TLMs were found to have better LM scores on words where recognition errors are more likely to occur. The R-TLM WER can be further reduced by interpolation with an LSTM-LM.

Recent speaker diarisation systems often convert variable length speech segments into fixed-length vector representations for speaker clustering, which are known as speaker embeddings. In this paper, the content-aware speaker embeddings (CASE) approach is proposed, which extends the input of the speaker classifier to include not only acoustic features but also their corresponding speech content, via phone, character, and word embeddings. Compared to alternative methods that leverage similar information, such as multitask or adversarial training, CASE factorises automatic speech recognition (ASR) from speaker recognition to focus on modelling speaker characteristics and correlations with the corresponding content units to derive more expressive representations. CASE is evaluated for speaker re-clustering with a realistic speaker diarisation setup using the AMI meeting transcription dataset, where the content information is obtained by performing ASR based on an automatic segmentation. Experimental results showed that CASE achieved a 17.8% relative speaker error rate reduction over conventional methods.

In this paper, a novel two-branch neural network model structure is proposed for multimodal emotion recognition, which consists of a time synchronous branch (TSB) and a time asynchronous branch (TAB). To capture correlations between each word and its acoustic realisation, the TSB combines speech and text modalities at each input window frame and then does pooling across time to form a single embedding vector. The TAB, by contrast, provides cross-utterance information by integrating sentence text embeddings from a number of context utterances into another embedding vector. The final emotion classification uses both the TSB and the TAB embeddings. Experimental results on the IEMOCAP dataset demonstrate that the two-branch structure achieves state-of-the-art results in 4-way classification with all common test setups. When using automatic speech recognition (ASR) output instead of manually transcribed reference text, it is shown that the cross-utterance information considerably improves the robustness against ASR errors. Furthermore, by incorporating an extra class for all the other emotions, the final 5-way classification system with ASR hypotheses can be viewed as a prototype for more realistic emotion recognition systems.

For various speech-related tasks, confidence scores from a speech recogniser are a useful measure to assess the quality of transcriptions. In traditional hidden Markov model-based automatic speech recognition (ASR) systems, confidence scores can be reliably obtained from word posteriors in decoding lattices. However, for an ASR system with an auto-regressive decoder, such as an attention-based sequence-to-sequence model, computing word posteriors is difficult. An obvious alternative is to use the decoder softmax probability as the model confidence. In this paper, we first examine how some commonly used regularisation methods influence the softmax-based confidence scores and study the overconfident behaviour of end-to-end models. Then we propose a lightweight and effective approach named confidence estimation module (CEM) on top of an existing end-to-end ASR model. Experiments on LibriSpeech show that CEM can mitigate the overconfidence problem and can produce more reliable confidence scores with and without shallow fusion of a language model. Further analysis shows that CEM generalises well to speech from a moderately mismatched domain and can potentially improve downstream tasks such as semi-supervised learning., Comment: Submitted to ICASSP 2021

The effective exploitation of richer contextual information in language models (LMs) is a long-standing research problem for automatic speech recognition (ASR). A cross-utterance LM (CULM) is proposed in this paper, which augments the input to a standard long short-term memory (LSTM) LM with a context vector derived from past and future utterances using an extraction network. The extraction network uses another LSTM to encode surrounding utterances into vectors which are integrated into a context vector using either a projection of LSTM final hidden states, or a multi-head self-attentive layer. In addition, an acoustic error sampling technique is proposed to reduce the mismatch between training and test-time. This is achieved by considering possible ASR errors into the model training procedure, and can therefore improve the word error rate (WER). Experiments performed on both AMI and Switchboard datasets show that CULMs outperform the LSTM LM baseline WER. In particular, the CULM with a self-attentive layer-based extraction network and acoustic error sampling achieves 0.6% absolute WER reduction on AMI, 0.3% WER reduction on the Switchboard part and 0.9% WER reduction on the Callhome part of Eval2000 test set over the respective baselines., Comment: 5 pages

In this paper, we propose a semi-supervised learning (SSL) technique for training deep neural networks (DNNs) to generate speaker-discriminative acoustic embeddings (speaker embeddings). Obtaining large amounts of speaker recognition train-ing data can be difficult for desired target domains, especially under privacy constraints. The proposed technique reduces requirements for labelled data by leveraging unlabelled data. The technique is a variant of virtual adversarial training (VAT) [1] in the form of a loss that is defined as the robustness of the speaker embedding against input perturbations, as measured by the cosine-distance. Thus, we term the technique cosine-distance virtual adversarial training (CD-VAT). In comparison to many existing SSL techniques, the unlabelled data does not have to come from the same set of classes (here speakers) as the labelled data. The effectiveness of CD-VAT is shown on the 2750+ hour VoxCeleb data set, where on a speaker verification task it achieves a reduction in equal error rate (EER) of 11.1% relative to a purely supervised baseline. This is 32.5% of the improvement that would be achieved from supervised training if the speaker labels for the unlabelled data were available., Comment: Accepted to Interspeech 2020

Speaker diarisation systems nowadays use embeddings generated from speech segments in a bottleneck layer, which are needed to be discriminative for unseen speakers. It is well-known that large-margin training can improve the generalisation ability to unseen data, and its use in such open-set problems has been widespread. Therefore, this paper introduces a general approach to the large-margin softmax loss without any approximations to improve the quality of speaker embeddings for diarisation. Furthermore, a novel and simple way to stabilise training, when large-margin softmax is used, is proposed. Finally, to combat the effect of overlapping speech, different training margins are used to reduce the negative effect overlapping speech has on creating discriminative embeddings. Experiments on the AMI meeting corpus show that the use of large-margin softmax significantly improves the speaker error rate (SER). By using all hyper parameters of the loss in a unified way, further improvements were achieved which reached a relative SER reduction of 24.6% over the baseline. However, by training overlapping and single speaker speech samples with different margins, the best result was achieved, giving overall a 29.5% SER reduction relative to the baseline., Comment: ICASSP 2020

In this paper, we propose Discriminative Neural Clustering (DNC) that formulates data clustering with a maximum number of clusters as a supervised sequence-to-sequence learning problem. Compared to traditional unsupervised clustering algorithms, DNC learns clustering patterns from training data without requiring an explicit definition of a similarity measure. An implementation of DNC based on the Transformer architecture is shown to be effective on a speaker diarisation task using the challenging AMI dataset. Since AMI contains only 147 complete meetings as individual input sequences, data scarcity is a significant issue for training a Transformer model for DNC. Accordingly, this paper proposes three data augmentation schemes: sub-sequence randomisation, input vector randomisation, and Diaconis augmentation, which generates new data samples by rotating the entire input sequence of L2-normalised speaker embeddings. Experimental results on AMI show that DNC achieves a reduction in speaker error rate (SER) of 29.4% relative to spectral clustering., Comment: Accepted as a conference paper at the 8th IEEE Spoken Language Technology Workshop (SLT 2021)

This paper proposes a novel automatic speech recognition (ASR) framework called Integrated Source-Channel and Attention (ISCA) that combines the advantages of traditional systems based on the noisy source-channel model (SC) and end-to-end style systems using attention-based sequence-to-sequence models. The traditional SC system framework includes hidden Markov models and connectionist temporal classification (CTC) based acoustic models, language models (LMs), and a decoding procedure based on a lexicon, whereas the end-to-end style attention-based system jointly models the whole process with a single model. By rescoring the hypotheses produced by traditional systems using end-to-end style systems based on an extended noisy source-channel model, ISCA allows structured knowledge to be easily incorporated via the SC-based model while exploiting the complementarity of the attention-based model. Experiments on the AMI meeting corpus show that ISCA is able to give a relative word error rate reduction up to 21% over an individual system, and by 13% over an alternative method which also involves combining CTC and attention-based models., Comment: To appear in Proc. ASRU2019, December 14-18, 2019, Sentosa, Singapore

This paper presents a new optimisation approach to train Deep Neural Networks (DNNs) with discriminative sequence criteria. At each iteration, the method combines information from the Natural Gradient (NG) direction with local curvature information of the error surface that enables better paths on the parameter manifold to be traversed. The method is derived using an alternative derivation of Taylor's theorem using the concepts of manifolds, tangent vectors and directional derivatives from the perspective of Information Geometry. The efficacy of the method is shown within a Hessian Free (HF) style optimisation framework to sequence train both standard fully-connected DNNs and Time Delay Neural Networks as speech recognition acoustic models. It is shown that for the same number of updates the proposed approach achieves larger reductions in the word error rate (WER) than both NG and HF, and also leads to a lower WER than standard stochastic gradient descent. The paper also addresses the issue of over-fitting due to mismatch between training criterion and Word Error Rate (WER) that primarily arises during sequence training of ReLU-DNN models., Comment: in Proc. INTERSPEECH 2018, September 2-6, 2018, Hyderabad, India

Deep Neural Network (DNN) acoustic models often use discriminative sequence training that optimises an objective function that better approximates the word error rate (WER) than frame-based training. Sequence training is normally implemented using Stochastic Gradient Descent (SGD) or Hessian Free (HF) training. This paper proposes an alternative batch style optimisation framework that employs a Natural Gradient (NG) approach to traverse through the parameter space. By correcting the gradient according to the local curvature of the KL-divergence, the NG optimisation process converges more quickly than HF. Furthermore, the proposed NG approach can be applied to any sequence discriminative training criterion. The efficacy of the NG method is shown using experiments on a Multi-Genre Broadcast (MGB) transcription task that demonstrates both the computational efficiency and the accuracy of the resulting DNN models., Comment: In Proceedings of IEEE ASRU 2017

This paper describes the extension and optimization of our previous work on very deep convolutional neural networks (CNNs) for effective recognition of noisy speech in the Aurora 4 task. The appropriate number of convolutional layers, the sizes of the filters, pooling operations and input feature maps are all modified: the filter and pooling sizes are reduced and dimensions of input feature maps are extended to allow adding more convolutional layers. Furthermore appropriate input padding and input feature map selection strategies are developed. In addition, an adaptation framework using joint training of very deep CNN with auxiliary features i-vector and fMLLR features is developed. These modifications give substantial word error rate reductions over the standard CNN used as baseline. Finally the very deep CNN is combined with an LSTM-RNN acoustic model and it is shown that state-level weighted log likelihood score combination in a joint acoustic model decoding scheme is very effective. On the Aurora 4 task, the very deep CNN achieves a WER of 8.81%, further 7.99% with auxiliary feature joint training, and 7.09% with LSTM-RNN joint decoding., Comment: accepted by SLT 2016