Speech representations learned in a self-supervised fashion from massive unlabeled speech corpora have been adapted successfully toward several downstream tasks. However, such representations may be skewed toward canonical data characteristics of such corpora and perform poorly on atypical, non-native accented speaker populations. With the state-of-the-art HuBERT model as a baseline, we propose and investigate self-supervised adaptation of speech representations to such populations in a parameter-efficient way via training accent-specific residual adapters. We experiment with 4 accents and choose automatic speech recognition (ASR) as the downstream task of interest. We obtain strong word error rate reductions (WERR) over HuBERT-large for all 4 accents, with a mean WERR of 22.7% with accent-specific adapters and a mean WERR of 25.1% if the entire encoder is accent-adapted. While our experiments utilize HuBERT and ASR as the downstream task, our proposed approach is both model and task-agnostic.
Conformer-based end-to-end models have become ubiquitous these days and are commonly used in both streaming and non-streaming automatic speech recognition (ASR). Techniques like dual-mode and dynamic chunk training helped unify streaming and non-streaming systems. However, there remains a performance gap between streaming with a full and limited past context. To address this issue, we propose the integration of a novel dynamic contextual carry-over mechanism in a state-of-the-art (SOTA) unified ASR system. Our proposed dynamic context Conformer (DCTX-Conformer) utilizes a non-overlapping contextual carry-over mechanism that takes into account both the left context of a chunk and one or more preceding context embeddings. We outperform the SOTA by a relative 25.0% word error rate, with a negligible latency impact due to the additional context embeddings.
End-to-end ASR models trained on large amount of data tend to be implicitly biased towards language semantics of the training data. Internal language model estimation (ILME) has been proposed to mitigate this bias for autoregressive models such as attention-based encoder-decoder and RNN-T. Typically, ILME is performed by modularizing the acoustic and language components of the model architecture, and eliminating the acoustic input to perform log-linear interpolation with the text-only posterior. However, for CTC-based ASR, it is not as straightforward to decouple the model into such acoustic and language components, as CTC log-posteriors are computed in a non-autoregressive manner. In this work, we propose a novel ILME technique for CTC-based ASR models. Our method iteratively masks the audio timesteps to estimate a pseudo log-likelihood of the internal LM by accumulating log-posteriors for only the masked timesteps. Extensive evaluation across multiple out-of-domain datasets reveals that the proposed approach improves WER by up to 9.8% and OOV F1-score by up to 24.6% relative to Shallow Fusion, when only text data from target domain is available. In the case of zero-shot domain adaptation, with no access to any target domain data, we demonstrate that removing the source domain bias with ILME can still outperform Shallow Fusion to improve WER by up to 9.3% relative.
Language models have been shown to perform better with an increase in scale on a wide variety of tasks via the in-context learning paradigm. In this paper, we investigate the hypothesis that the ability of a large language model to in-context learn-perform a task is not uniformly spread across all of its underlying components. Using a 66 billion parameter language model (OPT-66B) across a diverse set of 14 downstream tasks, we find this is indeed the case: $\sim$70% of attention heads and $\sim$20% of feed forward networks can be removed with minimal decline in task performance. We find substantial overlap in the set of attention heads (un)important for in-context learning across tasks and number of in-context examples. We also address our hypothesis through a task-agnostic lens, finding that a small set of attention heads in OPT-66B score highly on their ability to perform primitive induction operations associated with in-context learning, namely, prefix matching and copying. These induction heads overlap with task-specific important heads, suggesting that induction heads are among the heads capable of more sophisticated behaviors associated with in-context learning. Overall, our study provides several insights that indicate large language models may be under-trained to perform in-context learning and opens up questions on how to pre-train language models to more effectively perform in-context learning.
In this work, we define barge-in verification as a supervised learning task where audio-only information is used to classify user spoken dialogue into true and false barge-ins. Following the success of pre-trained models, we use low-level speech representations from a self-supervised representation learning model for our downstream classification task. Further, we propose a novel technique to infuse lexical information directly into speech representations to improve the domain-specific language information implicitly learned during pre-training. Experiments conducted on spoken dialog data show that our proposed model trained to validate barge-in entirely from speech representations is faster by 38% relative and achieves 4.5% relative F1 score improvement over a baseline LSTM model that uses both audio and Automatic Speech Recognition (ASR) 1-best hypotheses. On top of this, our best proposed model with lexically infused representations along with contextual features provides a further relative improvement of 5.7% in the F1 score but only 22% faster than the baseline.
End-to-end speech recognition models trained using joint Connectionist Temporal Classification (CTC)-Attention loss have gained popularity recently. In these models, a non-autoregressive CTC decoder is often used at inference time due to its speed and simplicity. However, such models are hard to personalize because of their conditional independence assumption that prevents output tokens from previous time steps to influence future predictions. To tackle this, we propose a novel two-way approach that first biases the encoder with attention over a predefined list of rare long-tail and out-of-vocabulary (OOV) words and then uses dynamic boosting and phone alignment network during decoding to further bias the subword predictions. We evaluate our approach on open-source VoxPopuli and in-house medical datasets to showcase a 60% improvement in F1 score on domain-specific rare words over a strong CTC baseline.
Although supervised deep learning has revolutionized speech and audio processing, it has necessitated the building of specialist models for individual tasks and application scenarios. It is likewise difficult to apply this to dialects and languages for which only limited labeled data is available. Self-supervised representation learning methods promise a single universal model that would benefit a wide variety of tasks and domains. Such methods have shown success in natural language processing and computer vision domains, achieving new levels of performance while reducing the number of labels required for many downstream scenarios. Speech representation learning is experiencing similar progress in three main categories: generative, contrastive, and predictive methods. Other approaches rely on multi-modal data for pre-training, mixing text or visual data streams with speech. Although self-supervised speech representation is still a nascent research area, it is closely related to acoustic word embedding and learning with zero lexical resources, both of which have seen active research for many years. This review presents approaches for self-supervised speech representation learning and their connection to other research areas. Since many current methods focus solely on automatic speech recognition as a downstream task, we review recent efforts on benchmarking learned representations to extend the application beyond speech recognition.
Automatic Speech Recognition (ASR) systems have found their use in numerous industrial applications in very diverse domains. Since domain-specific systems perform better than their generic counterparts on in-domain evaluation, the need for memory and compute-efficient domain adaptation is obvious. Particularly, adapting parameter-heavy transformer-based language models used for rescoring ASR hypothesis is challenging. In this work, we introduce domain-prompts, a methodology that trains a small number of domain token embedding parameters to prime a transformer-based LM to a particular domain. With just a handful of extra parameters per domain, we achieve 7-14% WER improvement over the baseline of using an unadapted LM. Despite being parameter-efficient, these improvements are comparable to those of fully-fine-tuned models with hundreds of millions of parameters. With ablations on prompt-sizes, dataset sizes, initializations and domains, we provide evidence for the benefits of using domain-prompts in ASR systems.
Many of the recent advances in speech separation are primarily aimed at synthetic mixtures of short audio utterances with high degrees of overlap. These datasets significantly differ from the real conversational data and hence, the models trained and evaluated on these datasets do not generalize to real conversational scenarios. Another issue with using most of these models for long form speech is the nondeterministic ordering of separated speech segments due to either unsupervised clustering for time-frequency masks or Permutation Invariant training (PIT) loss. This leads to difficulty in accurately stitching homogenous speaker segments for downstream tasks like Automatic Speech Recognition (ASR). In this paper, we propose a speaker conditioned separator trained on speaker embeddings extracted directly from the mixed signal. We train this model using a directed loss which regulates the order of the separated segments. With this model, we achieve significant improvements on Word error rate (WER) for real conversational data without the need for an additional re-stitching step.
Generic pre-trained speech and text representations promise to reduce the need for large labeled datasets on specific speech and language tasks. However, it is not clear how to effectively adapt these representations for speech emotion recognition. Recent public benchmarks show the efficacy of several popular self-supervised speech representations for emotion classification. In this study, we show that the primary difference between the top-performing representations is in predicting valence while the differences in predicting activation and dominance dimensions are less pronounced. However, we show that even the best-performing HuBERT representation underperforms on valence prediction compared to a multimodal model that also incorporates text representation. We address this shortcoming by injecting lexical information into the speech representation using the multimodal model as a teacher. To improve the efficacy of our approach, we propose a novel estimate of the quality of the emotion predictions, to condition teacher-student training. We report new audio-only state-of-the-art concordance correlation coefficient (CCC) values of 0.757, 0.627, 0.671 for activation, valence and dominance predictions, respectively, on the MSP-Podcast corpus, and also state-of-the-art values of 0.667, 0.582, 0.545 on the IEMOCAP corpus.