In the realm of spoken language understanding (SLU), numerous natural language understanding (NLU) methodologies have been adapted by supplying large language models (LLMs) with transcribed speech instead of conventional written text. In real-world scenarios, prior to input into an LLM, an automated speech recognition (ASR) system generates an output transcript hypothesis, where inherent errors can degrade subsequent SLU tasks. Here we introduce a method that utilizes the ASR system's lattice output instead of relying solely on the top hypothesis, aiming to encapsulate speech ambiguities and enhance SLU outcomes. Our in-context learning experiments, covering spoken question answering and intent classification, underline the LLM's resilience to noisy speech transcripts with the help of word confusion networks from lattices, bridging the SLU performance gap between using the top ASR hypothesis and an oracle upper bound. Additionally, we delve into the LLM's robustness to varying ASR performance conditions and scrutinize the aspects of in-context learning which prove the most influential.
The area of Machine Learning as a Service (MLaaS) is experiencing increased implementation due to recent advancements in the AI (Artificial Intelligence) industry. However, this spike has prompted concerns regarding AI defense mechanisms, specifically regarding potential covert attacks from third-party providers that cannot be entirely trusted. Recent research has uncovered that auditory backdoors may use certain modifications as their initiating mechanism. DynamicTrigger is introduced as a methodology for carrying out dynamic backdoor attacks that use cleverly designed tweaks to ensure that corrupted samples are indistinguishable from clean. By utilizing fluctuating signal sampling rates and masking speaker identities through dynamic sound triggers (such as the clapping of hands), it is possible to deceive speech recognition systems (ASR). Our empirical testing demonstrates that DynamicTrigger is both potent and stealthy, achieving impressive success rates during covert attacks while maintaining exceptional accuracy with non-poisoned datasets.
Automatic speech recognition (ASR) systems can suffer from poor recall for various reasons, such as noisy audio, lack of sufficient training data, etc. Previous work has shown that recall can be improved by retrieving rewrite candidates from a large database of likely, contextually-relevant alternatives to the hypothesis text using nearest-neighbors search over embeddings of the ASR hypothesis text to correct and candidate corrections. However, ASR-hypothesis-based retrieval can yield poor precision if the textual hypotheses are too phonetically dissimilar to the transcript truth. In this paper, we eliminate the hypothesis-audio mismatch problem by querying the correction database directly using embeddings derived from the utterance audio; the embeddings of the utterance audio and candidate corrections are produced by multimodal speech-text embedding networks trained to place the embedding of the audio of an utterance and the embedding of its corresponding textual transcript close together. After locating an appropriate correction candidate using nearest-neighbor search, we score the candidate with its speech-text embedding distance before adding the candidate to the original n-best list. We show a relative word error rate (WER) reduction of 6% on utterances whose transcripts appear in the candidate set, without increasing WER on general utterances.
The objective of sign language recognition is to bridge the communication gap between the deaf and the hearing. Numerous previous works train their models using the well-established connectionist temporal classification (CTC) loss. During the inference stage, the CTC-based models typically take the entire sign video as input to make predictions. This type of inference scheme is referred to as offline recognition. In contrast, while mature speech recognition systems can efficiently recognize spoken words on the fly, sign language recognition still falls short due to the lack of practical online solutions. In this work, we take the first step towards filling this gap. Our approach comprises three phases: 1) developing a sign language dictionary encompassing all glosses present in a target sign language dataset; 2) training an isolated sign language recognition model on augmented signs using both conventional classification loss and our novel saliency loss; 3) employing a sliding window approach on the input sign sequence and feeding each sign clip to the well-optimized model for online recognition. Furthermore, our online recognition model can be extended to boost the performance of any offline model, and to support online translation by appending a gloss-to-text network onto the recognition model. By integrating our online framework with the previously best-performing offline model, TwoStream-SLR, we achieve new state-of-the-art performance on three benchmarks: Phoenix-2014, Phoenix-2014T, and CSL-Daily. Code and models will be available at https://github.com/FangyunWei/SLRT
Speech Emotion Recognition (SER) has become a growing focus of research in human-computer interaction. Spatiotemporal features play a crucial role in SER, yet current research lacks comprehensive spatiotemporal feature learning. This paper focuses on addressing this gap by proposing a novel approach. In this paper, we employ Convolutional Neural Network (CNN) with varying kernel sizes for spatial and temporal feature extraction. Additionally, we introduce Squeeze-and-Excitation (SE) modules to capture and fuse multi-scale features, facilitating effective information fusion for improved emotion recognition and a deeper understanding of the temporal evolution of speech emotion. Moreover, we employ skip connections and Spatial Dropout (SD) layers to prevent overfitting and increase the model's depth. Our method outperforms the previous state-of-the-art method, achieving an average UAR and WAR improvement of 1.62% and 1.32%, respectively, across six benchmark SER datasets. Further experiments demonstrated that our method can fully extract spatiotemporal features in low-resource conditions.
One of the major challenges for developing automatic speech recognition (ASR) for low-resource languages is the limited access to labeled data with domain-specific variations. In this study, we propose a pseudo-labeling approach to develop a large-scale domain-agnostic ASR dataset. With the proposed methodology, we developed a 20k+ hours labeled Bangla speech dataset covering diverse topics, speaking styles, dialects, noisy environments, and conversational scenarios. We then exploited the developed corpus to design a conformer-based ASR system. We benchmarked the trained ASR with publicly available datasets and compared it with other available models. To investigate the efficacy, we designed and developed a human-annotated domain-agnostic test set composed of news, telephony, and conversational data among others. Our results demonstrate the efficacy of the model trained on psuedo-label data for the designed test-set along with publicly-available Bangla datasets. The experimental resources will be publicly available.(https://github.com/hishab-nlp/Pseudo-Labeling-for-Domain-Agnostic-Bangla-ASR)
Whispering is a distinct form of speech known for its soft, breathy, and hushed characteristics, often used for private communication. The acoustic characteristics of whispered speech differ substantially from normally phonated speech and the scarcity of adequate training data leads to low automatic speech recognition (ASR) performance. To address the data scarcity issue, we use a signal processing-based technique that transforms the spectral characteristics of normal speech to those of pseudo-whispered speech. We augment an End-to-End ASR with pseudo-whispered speech and achieve an 18.2% relative reduction in word error rate for whispered speech compared to the baseline. Results for the individual speaker groups in the wTIMIT database show the best results for US English. Further investigation showed that the lack of glottal information in whispered speech has the largest impact on whispered speech ASR performance.
Automatic speech recognition (ASR) has reached a level of accuracy in recent years, that even outperforms humans in transcribing speech to text. Nevertheless, all current ASR approaches show a certain weakness against ambient noise. To reduce this weakness, audio-visual speech recognition (AVSR) approaches additionally consider visual information from lip movements for transcription. This additional modality increases the computational cost for training models from scratch. We propose an approach, that builds on a pre-trained ASR model and extends it with an adaptive upstream module, that fuses audio and visual information. Since we do not need to train the transformer structure from scratch, our approach requires a fraction of the computational resources compared to traditional AVSR models. Compared to current SOTA systems like AV-HuBERT, our approach achieves an average improvement of 8.3% in word error rate across different model sizes, noise categories and broad SNR range. The approach allows up to 21% smaller models and requires only a fraction of the computational resources for training and inference compared to common AVSR approaches.
Confidence estimation of predictions from an End-to-End (E2E) Automatic Speech Recognition (ASR) model benefits ASR's downstream and upstream tasks. Class-probability-based confidence scores do not accurately represent the quality of overconfident ASR predictions. An ancillary Confidence Estimation Model (CEM) calibrates the predictions. State-of-the-art (SOTA) solutions use binary target scores for CEM training. However, the binary labels do not reveal the granular information of predicted words, such as temporal alignment between reference and hypothesis and whether the predicted word is entirely incorrect or contains spelling errors. Addressing this issue, we propose a novel Temporal-Lexeme Similarity (TeLeS) confidence score to train CEM. To address the data imbalance of target scores while training CEM, we use shrinkage loss to focus on hard-to-learn data points and minimise the impact of easily learned data points. We conduct experiments with ASR models trained in three languages, namely Hindi, Tamil, and Kannada, with varying training data sizes. Experiments show that TeLeS generalises well across domains. To demonstrate the applicability of the proposed method, we formulate a TeLeS-based Acquisition (TeLeS-A) function for sampling uncertainty in active learning. We observe a significant reduction in the Word Error Rate (WER) as compared to SOTA methods.
In a spoken dialogue system, an NLU model is preceded by a speech recognition system that can deteriorate the performance of natural language understanding. This paper proposes a method for investigating the impact of speech recognition errors on the performance of natural language understanding models. The proposed method combines the back transcription procedure with a fine-grained technique for categorizing the errors that affect the performance of NLU models. The method relies on the usage of synthesized speech for NLU evaluation. We show that the use of synthesized speech in place of audio recording does not change the outcomes of the presented technique in a significant way.