Recent works have shown that prompting large language models with audio encodings can unlock speech recognition capabilities. However, existing techniques do not scale efficiently, especially while handling long form streaming audio inputs — not only do they extrapolate poorly beyond the audio length seen during training, but they are also computationally inefficient due to the quadratic cost of attention.In this work, we introduce SpeechLLM-XL, a linear scaling decoder-only model for streaming speech recognition. We process audios in configurable chunks using limited attention window for reduced computation, and the text tokens for each audio chunk are generated auto-regressively until an EOS is predicted. During training, the transcript is segmented into chunks, using a CTC forced alignment estimated from encoder output. SpeechLLM-XL with 1.28 seconds chunk size achieves 2.7%/6.7% WER on LibriSpeech test clean/other, and it shows no quality degradation on long form utterances 10x longer than the training utterances.

Large language models (LLMs) have become proficient at solving a wide variety of tasks, including those involving multi-modal inputs. In particular, instantiating an LLM (such as LLaMA) with a speech encoder and training it on paired data imparts speech recognition (ASR) abilities to the decoder-only model, hence called Speech-LLaMA. Nevertheless, due to the sequential nature of auto-regressive inference and the relatively large decoder, Speech-LLaMA models require relatively high inference time. In this work, we propose to speed up Speech-LLaMA inference by predicting multiple tokens in the same decoding step. We explore several model architectures that enable this, and investigate their performance using threshold-based and verification-based inference strategies. We also propose a prefix-based beam search decoding method that allows efficient minimum word error rate (MWER) training for such models. We evaluate our models on a variety of public benchmarks, where they reduce the number of decoder calls by ∼3.2x while maintaining or improving WER performance.

Speech LLMs use speech embeddings as the prompt to a Large Language Model (LLM) and generate human readable text for the speech signal in an autoregressive manner. Teacher-forcing is a common approach used for training Speech LLMs, which is dissimilar to the procedure used during inference, creating a gap between training and inference known as exposure bias. To mitigate exposure bias, we propose Speech-N-LlaMA. Contrary to the existing Speech LLMs that have single pass through the LLM during training, Speech-N-LlaMA incorporates multi-pass training. Through multiple passes, Speech-N-LlaMA mitigates exposure bias and uses the error correction capability of LLM to improve the performance of Speech LLMs. We propose an N-pass loss and utterance level temperature sampling in Speech-N-LlaMA to achieve this. We evaluate four different model sizes on three benchmarks, and show up to 18% relative improvement in Word Error Rate (WER) compared to baseline Speech LLM, while not incurring any additional compute during inference.

The growing popularity of multi-channel wearable devices, such as smart glasses, has led to a surge of applications such as targeted speech recognition and enhanced hearing. However, current approaches to solve these tasks use independently trained models, which may not benefit from large amounts of unlabeled data. In this paper, we propose M-BEST-RQ, the first multi-channel speech foundation model for smart glasses, which is designed to leverage large-scale self-supervised learning (SSL) in an array-geometry agnostic approach. While prior work on multi-channel speech SSL only evaluated on simulated settings, we curate a suite of real downstream tasks to evaluate our model, namely (i) conversational automatic speech recognition (ASR), (ii) spherical active source localization, and (iii) glasses wearer voice activity detection, which are sourced from the MMCSG and EasyCom datasets. We show that a general-purpose M-BEST-RQ encoder is able to match or surpass supervised models across all tasks. For the conversational ASR task in particular, using only 8 hours of labeled speech, our model outperforms a supervised ASR baseline that is trained on 2000 hours of labeled data, which demonstrates the effectiveness of our approach.

CTC compressor can be an effective approach to integrate audio encoders to decoder-only models, which has gained growing interest for different speech applications. In this work, we propose a novel CTC compressor based joint speech and text training (CJST) framework for decoder-only ASR. CJST matches speech and text modalities from both directions by exploring a simple modality adaptor and several features of the CTC compressor, including sequence compression, on-the-fly forced peaky alignment and CTC class embeddings. Experimental results on the Librispeech and TED-LIUM2 corpora show that the proposed CJST achieves an effective text injection without the need of duration handling, leading to the best performance for both in-domain and cross-domain scenarios. We also provide a comprehensive study on CTC compressor, covering various compression modes, edge case handling and behavior under both clean and noisy data conditions, which reveals the most robust setting to use CTC compressor for decoder-only models.

Large language models (LLMs) have proven themselves highly flexible, able to solve a wide range of generative tasks, such as abstractive summarization and open-ended question answering. In this paper we extend the capabilities of LLM by directly attaching a small audio encoder allowing it to perform speech recognition. By directly prepending a sequence of audio embeddings to the text token embeddings, the LLM can be converted to an automatic speech recognition (ASR) system, and be used in the exact same manner as its textual counterpart. Experiments on Multilingual LibriSpeech (MLS) show that incorporating a conformer encoder into the open sourced LLaMA-7B allows it to outperform monolingual baselines by 18% relatively in WER and perform multilingual speech recognition, despite LLaMA being trained overwhelmingly on English text. Furthermore, we perform ablation studies to investigate whether the LLM can be completely frozen during training to maintain its original capabilities, scaling up the audio encoder, and increasing the audio encoder striding to generate fewer embeddings. The results from these studies show that multilingual ASR is possible even when the LLM is frozen, or when strides of almost 1 second are used in the audio encoder opening up the possibility for LLMs to operate on long-form audio.

Automatic Speech Recognition (ASR) models need to be optimized for specific hardware before they can be deployed on devices. This can be done by tuning the model’s hyperparameters or exploring variations in its architecture. Re-training and re-validating models after making these changes can be a resource-intensive task. This paper presents TODM (Train Once Deploy Many), a new approach to efficiently train many sizes of hardware-friendly on-device ASR models with comparable GPU-hours to that of a single training job. TODM leverages insights from prior work on Supernet, where Recurrent Neural Network Transducer (RNN-T) models share weights within a Supernet. It reduces layer sizes and widths of the Supernet to obtain subnetworks, making them smaller models suitable for all hardware types. We introduce a novel combination of three techniques to improve the outcomes of the TODM Supernet: adaptive dropout, an in-place Alpha-divergence knowledge distillation, and the use of ScaledAdam optimizer. We validate our approach by comparing Supernet-trained versus individually tuned Multi-Head State Space Model (MH-SSM) RNN-T using LibriSpeech. Results demonstrate that our TODM Supernet either matches or surpasses the performance of manually tuned models by up to a relative of 3% better in word error rate (WER), while efficiently keeping the cost of training many models at a small constant.

Neural network pruning offers an effective method for compressing a multilingual automatic speech recognition (ASR) model with minimal performance loss. However, it entails several rounds of pruning and re-training needed to be run for each language. In this work, we propose the use of an adaptive masking approach in two scenarios for pruning a multilingual ASR model efficiently, each resulting in sparse monolingual models or a sparse multilingual model (named as Dynamic ASR Pathways). Our approach dynamically adapts the subnetwork, avoiding premature decisions about a fixed sub-network structure. We show that our approach outperforms existing pruning methods when targeting sparse monolingual models. Further, we illustrate that Dynamic ASR Pathways jointly discovers and trains better sub-networks (pathways) of a single multilingual model by adapting from different sub-network initializations, thereby reducing the need for language-specific pruning.

Yassir Fathullah, Chunyang Wu, Egor Lakomkin, Ke Li, Junteng Jia, Yuan Shangguan, Jay Mahadeokar, Ozlem Kalinli, Christian Fuegen, Mike Seltzer. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

In this paper, we propose a joint federated learning (FL) and personalization method for on-device ASR adaptation. Starting with a Conformer-based RNN-T as the ASR model backbone that is pretrained on public data and shared across devices, we propose to adapt to user data on-device by ① collectively finetuning the backbone on all user data by FL, and ② for each user, augmenting the backbone with a personalized adapter that is trained on on-device data and stored locally on their devices. As ground-truth transcriptions are not available, we use pseudo-label training, which can be completely performed on-device. Our joint method combines the best of both FL and personalization and achieves maximum effect for both heavy and light users: For users with 50+ adaptation utterances, our recipe achieves $27.4 \%$ relative WER reduction, largely due to personalization; for users with no adaptation utterance, our recipe achieves $8.9 \%$ relative WER reduction purely due to FL.

Neural transducers have achieved human level performance on standard speech recognition benchmarks. However, their performance significantly degrades in the presence of cross-talk, especially when the primary speaker has a low signal-to-noise ratio. Anchored speech recognition refers to a class of methods that use information from an anchor segment (e.g., wake-words) to recognize device-directed speech while ignoring interfering background speech. In this paper, we investigate anchored speech recognition to make neural transducers robust to background speech. We extract context information from the anchor segment with a tiny auxiliary network, and use encoder biasing and joiner gating to guide the transducer towards the target speech. Moreover, to improve the robustness of context embedding extraction, we propose auxiliary training objectives to disentangle lexical content from speaking style. We evaluate our methods on synthetic LibriSpeech-based mixtures comprising several SNR and overlap conditions; they improve relative word error rates by 19.6% over a strong baseline, when averaged over all conditions.

Graph neural networks (GNNs) have shown superiority in many prediction tasks over graphs due to their impressive capability of capturing nonlinear relations in graph-structured data. However, for node classification tasks, often, only marginal improvement of GNNs has been observed in practice over their linear counterparts. Previous works provide very few understandings of this phenomenon. In this work, we resort to Bayesian learning to give an in-depth investigation of the functions of non-linearity in GNNs for node classification tasks. Given a graph generated from the statistical model CSBM, we observe that the max-a-posterior estimation of a node label given its own and neighbors' attributes consists of two types of non-linearity, the transformation of node attributes and a ReLU-activated feature aggregation from neighbors. The latter surprisingly matches the type of non-linearity used in many GNN models. By further imposing Gaussian assumption on node attributes, we prove that the superiority of those ReLU activations is only significant when the node attributes are far more informative than the graph structure, which nicely explains previous empirical observations. A similar argument is derived when there is a distribution shift of node attributes between the training and testing datasets. Finally, we verify our theory on both synthetic and real-world networks. Our code is available at https://github.com/Graph-COM/Bayesian_inference_based_GNN.git.

A graph neural network transforms features in each vertex's neighborhood into a vector representation of the vertex. Afterward, each vertex's representation is used independently for predicting its label. This standard pipeline implicitly assumes that vertex labels are conditionally independent given their neighborhood features. However, this is a strong assumption, and we show that it is far from true on many real-world graph datasets. Focusing on regression tasks, we find that this conditional independence assumption severely limits predictive power. This should not be that surprising, given that traditional graph-based semi-supervised learning methods such as label propagation work in the opposite fashion by explicitly modeling the correlation in predicted outcomes.

We present a graph-based semi-supervised learning (SSL) method for learning edge flows defined on a graph. Specifically, given flow measurements on a subset of edges, we want to predict the flows on the remaining edges. To this end, we develop a computational framework that imposes certain constraints on the overall flows, such as (approximate) flow conservation. These constraints render our approach different from classical graph-based SSL for vertex labels, which posits that tightly connected nodes share similar labels and leverages the graph structure accordingly to extrapolate from a few vertex labels to the unlabeled vertices. We derive bounds for our method's reconstruction error and demonstrate its strong performance on synthetic and real-world flow networks from transportation, physical infrastructure, and the Web. Furthermore, we provide two active learning algorithms for selecting informative edges on which to measure flow, which has applications for optimal sensor deployment. The first strategy selects edges to minimize the reconstruction error bound and works well on flows that are approximately divergence-free. The second approach clusters the graph and selects bottleneck edges that cross cluster-boundaries, which works well on flows with global trends.

Many time series are effectively generated by a combination of deterministic continuous flows along with discrete jumps sparked by stochastic events. However, we usually do not have the equation of motion describing the flows, or how they are affected by jumps. To this end, we introduce Neural Jump Stochastic Differential Equations that provide a data-driven approach to learn continuous and discrete dynamic behavior, i.e., hybrid systems that both flow and jump. Our approach extends the framework of Neural Ordinary Differential Equations with a stochastic process term that models discrete events. We then model temporal point processes with a piecewise-continuous latent trajectory, where the discontinuities are caused by stochastic events whose conditional intensity depends on the latent state. We demonstrate the predictive capabilities of our model on a range of synthetic and real-world marked point process datasets, including classical point processes (such as Hawkes processes), awards on Stack Overflow, medical records, and earthquake monitoring.

Core-periphery structure is a common property of complex networks, which is a composition of tightly connected groups of core vertices and sparsely connected periphery vertices. This structure frequently emerges in traffic systems, biology, and social networks via underlying spatial positioning of the vertices. While core-periphery structure is ubiquitous, there have been limited attempts at modeling network data with this structure. Here, we develop a generative, random network model with core-periphery structure that jointly accounts for topological and spatial information by "core scores'' of vertices. Our model achieves substantially higher likelihood than existing generative models of core-periphery structure, and we demonstrate how the core scores can be used in downstream data mining tasks, such as predicting airline traffic and classifying fungal networks. We also develop nearly linear time algorithms for learning model parameters and network sampling by using a method akin to the fast multipole method, a technique traditional to computational physics, which allow us to scale to networks with millions of vertices with minor tradeoffs in accuracy.