Multilingual ASR offers training, deployment and overall performance benefits, but models trained via simple data pooling are known to suffer from cross-lingual interference. Oracle language information (exact-prior) and language-specific parameters are usually leveraged to overcome this, but such approaches cannot enable seamless, truly multilingual experiences. Existing methods try to overcome this limitation by relying on inferred language information or language agnostic mixture-of-experts, but they incur additional runtime complexity and/or training cost in addition to being less effective in streaming scenarios. Building on previous studies where models were trained to handle mixed-prior (knowledge that the underlying language belongs to a known group), we propose Attentive Multilingual Speech Encoding (AMuSE), a training framework designed to match exact-prior performance even in the absence of underlying language information at runtime (zero-prior), thereby making the model prior-agnostic. Leveraging AMuSE, we build a zero-prior enabled LLM-based ASR system that outperforms several exact-prior driven state-of-the-art benchmarks.

Dialog systems, such as voice assistants, are expected to engage with users in complex, evolving conversations. Unfortunately, traditional automatic speech recognition (ASR) systems deployed in such applications are usually trained to recognize each turn independently and lack the ability to adapt to the conversational context or incorporate user feedback. In this work, we introduce a general framework for ASR in dialog systems that can go beyond learning from single-turn utterances and learn over time how to adapt to both explicit supervision and implicit user feedback present in multi-turn conversations. We accomplish that by leveraging advances in student-teacher learning and context-aware dialog processing, and designing contrastive self-supervision approaches with Ohm, a new online hard-negative mining approach. We show that leveraging our new framework compared to traditional training leads to relative WER reductions of close to 10% in real-world dialog systems, and up to 26% on public synthetic data.

Conventional dynamic language switching enables seamless multilingual interactions by running several monolingual ASR systems in parallel and triggering the appropriate downstream components using a standalone language identification (LID) service. Since this solution is neither scalable nor cost- and memory-efficient, especially for on-device applications, we propose end-to-end, streaming, joint ASR-LID architectures based on the recurrent neural network transducer framework. Two key formulations are explored: (1) joint training using a unified output space for ASR and LID vocabularies, and (2) joint training viewed as multi-task optimization. We also evaluate the benefit of using auxiliary language information obtained on-the-fly from an acoustic LID classifier. Experiments with the English-Hindi language pair show that: (a) multi-task architectures perform better overall, and (b) the best joint architecture surpasses monolingual ASR (6.4–9.2% word error rate reduction) and acoustic LID (53.9–56.1% error rate reduction) baselines while reducing the overall memory footprint by up to 46%.

Semi-supervised learning (SSL) is an active area of research which aims to utilize unlabeled data to improve the accuracy of speech recognition systems. While the previous studies have established the efficacy of various SSL methods on varying amounts of data, this paper presents largest ASR SSL experiment ever conducted till date where 75K hours of labeled and 1.2 million hours of unlabeled data is used for model training. In addition, the paper introduces couple of novel techniques to facilitate such a large scale experiment: 1) a simple scalable Teacher-Student based SSL method for connectionist temporal classification (CTC) objective and 2) effective data selection mechanisms for leveraging massive amounts of unlabeled data to boost the performance of student models. Further, we apply SSL in all stages of the acoustic model training, including final stage sequence discriminative training. Our experiments indicate encouraging word error rate (WER) gains up to 14% in such a large transcribed data regime due to the SSL training.

Building conversational speech recognition systems for new languages is constrained by the availability of utterances capturing user-device interactions. Data collection is expensive and limited by speed of manual transcription. In order to address this, we advocate the use of neural machine translation as a data augmentation technique for bootstrapping language models. Machine translation (MT) offers a systematic way of incorporating collections from mature, resource-rich conversational systems that may be available for a different language. However, ingesting raw translations from a general purpose MT system may not be effective owing to the presence of named entities, intra sentential code-switching and the domain mismatch between the conversational data being translated and the parallel text used for MT training. To circumvent this, we explore following domain adaptation techniques: (a) sentence embedding based data selection for MT training, (b) model finetuning, and (c) rescoring and filtering translated hypotheses. Using Hindi language as the experimental testbed, we supplement transcribed collections with translated US English utterances. We observe a relative word error rate reduction of 7.8-15.6%, depending on the bootstrapping phase. Fine grained analysis reveals that translation particularly aids the interaction scenarios underrepresented in the transcribed data.

In the past, conventional i-vectors based on a Universal Background Model (UBM) have been successfully used as input features to adapt a Deep Neural Network (DNN) Acoustic Model (AM) for Automatic Speech Recognition (ASR). In contrast, this paper introduces Hidden Markov Model (HMM) based ivectors that use HMM state alignment information from an ASR system for estimating i-vectors. Further, we propose passing these HMM based i-vectors though an explicit non-linear hidden layer of a DNN before combining them with standard acoustic features, such as log filter bank energies (LFBEs). To improve robustness to mismatched adaptation data, we also propose estimating i-vectors in a causal fashion for training the DNN, restricting the connectivity among hidden nodes in the DNN and applying a max-pool non-linearity at selected hidden nodes. In our experiments, these techniques yield about 5-7% relative word error rate (WER) improvement over the baseline speaker independent system in matched condition, and a substantial WER reduction for mismatched adaptation data.

We investigate the problem of speaker adaptation of DNN acoustic models in two settings: the traditional unsupervised adaptation and a supervised adaptation (SuA) where a few minutes of transcribed speech is available. SuA presents additional difficulties when a test speaker’s adaptation information does not match the registered speaker’s information. Employing feature-space maximum likelihood linear regression (fMLLR) transformed features as side-information to the DNN, we reintroduce some classical ideas for combining adapted and unadapted features: early and late fusion methods, as well as the estimation of the fMLLR transforms using simple target models (STM). Results show that early fusion helps DNNs generalize better when features are combined after a non-linear bottleneck layer, while late fusion improves robustness, specifically in mismatched cases. STM give consistent improvements in both settings.

Auto-associative neural network (AANN) is a fully connected feed-forward neural network, trained to reconstruct its input at its output through a hidden compression layer, which has fewer numbers of nodes than the dimensionality of input. AANNs are used to model speakers in speaker verification, where a speaker-specific AANN model is obtained by adapting (or retraining) the universal background model (UBM) AANN, an AANN trained on multiple held out speakers, using corresponding speaker data. When the amount of speaker data is limited, this adaptation procedure may lead to overfitting as all the parameters of UBM-AANN are adapted. In this paper, we introduce and develop the factor analysis theory of AANNs to alleviate this problem. We hypothesize that only the weight matrix connecting the last nonlinear hidden layer and the output layer is speaker-specific, and further restrict it to a common low-dimensional subspace during adaptation. The subspace is learned using large amounts of development data, and is held fixed during adaptation. Thus, only the coordinates in a subspace, also known as i-vector, need to be estimated using speaker-specific data. The update equations are derived for learning both the common low-dimensional subspace and the i-vectors corresponding to speakers in the subspace. The resultant i-vector representation is used as a feature for the probabilistic linear discriminant analysis model. The proposed system shows promising results on the NIST-08 speaker recognition evaluation (SRE), and yields a 23% relative improvement in equal error rate over the previously proposed weighted least squares-based subspace AANNs system. The experiments on NIST-10 SRE confirm that these improvements are consistent and generalize across datasets.

Auto-Associative Neural Network (AANN) is a fully connected feed-forward neural network, trained to reconstruct its input at its output through a hidden compression layer. AANNs are used to model speakers in speaker verification, where a speaker-specific AANN model is obtained by adapting (or retraining) the Universal Background Model (UBM) AANN, an AANN trained on multiple held out speakers, using corresponding speaker data. When the amount of speaker data is limited, this adaptation procedure leads to overfitting. Additionally, the resultant speaker-specific parameters become noisy due to outliers in data. Thus, we propose to regularize the parameters of an AANN during speaker adaptation. A closed-form expression for updating the parameters is derived. Further, these speaker-specific AANN parameters are directly used as features in linear discriminant analysis (LDA)/probabilistic discriminant (PLDA) analysis based speaker verification system. The proposed speaker verification system outperforms the previously proposed weighted least squares (WLS) based AANN speaker verification system on NIST-08 speaker recognition evaluation (SRE). Moreover, the proposed speaker verification system obviates the need for an intermediate dimensionality reduction (or i-vector extraction) step.