Siddhant Arora

Yifan Peng, Jinchuan Tian, Brian Yan et al. · cmu, meta-ai

Pre-training speech models on large volumes of data has achieved remarkable success. OpenAI Whisper is a multilingual multitask model trained on 680k hours of supervised speech data. It generalizes well to various speech recognition and translation benchmarks even in a zero-shot setup. However, the full pipeline for developing such models (from data collection to training) is not publicly accessible, which makes it difficult for researchers to further improve its performance and address training-related issues such as efficiency, robustness, fairness, and bias. This work presents an Open Whisper-style Speech Model (OWSM), which reproduces Whisper-style training using an open-source toolkit and publicly available data. OWSM even supports more translation directions and can be more efficient to train. We will publicly release all scripts used for data preparation, training, inference, and scoring as well as pre-trained models and training logs to promote open science.

Chien-yu Huang, Ke-Han Lu, Shih-Heng Wang et al. · cmu, meta-ai

Text language models have shown remarkable zero-shot capability in generalizing to unseen tasks when provided with well-formulated instructions. However, existing studies in speech processing primarily focus on limited or specific tasks. Moreover, the lack of standardized benchmarks hinders a fair comparison across different approaches. Thus, we present Dynamic-SUPERB, a benchmark designed for building universal speech models capable of leveraging instruction tuning to perform multiple tasks in a zero-shot fashion. To achieve comprehensive coverage of diverse speech tasks and harness instruction tuning, we invite the community to collaborate and contribute, facilitating the dynamic growth of the benchmark. To initiate, Dynamic-SUPERB features 55 evaluation instances by combining 33 tasks and 22 datasets. This spans a broad spectrum of dimensions, providing a comprehensive platform for evaluation. Additionally, we propose several approaches to establish benchmark baselines. These include the utilization of speech models, text language models, and the multimodal encoder. Evaluation results indicate that while these baselines perform reasonably on seen tasks, they struggle with unseen ones. We release all materials to the public and welcome researchers to collaborate on the project, advancing technologies in the field together.

Suwon Shon, Siddhant Arora, Chyi-Jiunn Lin et al. · cmu, deepmind

Spoken language understanding (SLU) tasks have been studied for many decades in the speech research community, but have not received as much attention as lower-level tasks like speech and speaker recognition. In particular, there are not nearly as many SLU task benchmarks, and many of the existing ones use data that is not freely available to all researchers. Recent work has begun to introduce such benchmark datasets for several tasks. In this work, we introduce several new annotated SLU benchmark tasks based on freely available speech data, which complement existing benchmarks and address gaps in the SLU evaluation landscape. We contribute four tasks: question answering and summarization involve inference over longer speech sequences; named entity localization addresses the speech-specific task of locating the targeted content in the signal; dialog act classification identifies the function of a given speech utterance. We follow the blueprint of the Spoken Language Understanding Evaluation (SLUE) benchmark suite. In order to facilitate the development of SLU models that leverage the success of pre-trained speech representations, we will be publishing for each task (i) annotations for a relatively small fine-tuning set, (ii) annotated development and test sets, and (iii) baseline models for easy reproducibility and comparisons. In this work, we present the details of data collection and annotation and the performance of the baseline models. We also perform sensitivity analysis of pipeline models' performance (speech recognizer + text model) to the speech recognition accuracy, using more than 20 state-of-the-art speech recognition models.

Yifan Peng, Siddhant Arora, Yosuke Higuchi et al. · cmu, nvidia

Collecting sufficient labeled data for spoken language understanding (SLU) is expensive and time-consuming. Recent studies achieved promising results by using pre-trained models in low-resource scenarios. Inspired by this, we aim to ask: which (if any) pre-training strategies can improve performance across SLU benchmarks? To answer this question, we employ four types of pre-trained models and their combinations for SLU. We leverage self-supervised speech and language models (LM) pre-trained on large quantities of unpaired data to extract strong speech and text representations. We also explore using supervised models pre-trained on larger external automatic speech recognition (ASR) or SLU corpora. We conduct extensive experiments on the SLU Evaluation (SLUE) benchmark and observe self-supervised pre-trained models to be more powerful, with pre-trained LM and speech models being most beneficial for the Sentiment Analysis and Named Entity Recognition task, respectively.

Siddhant Arora, Siddharth Dalmia, Xuankai Chang et al. · cmu

End-to-end (E2E) models are becoming increasingly popular for spoken language understanding (SLU) systems and are beginning to achieve competitive performance to pipeline-based approaches. However, recent work has shown that these models struggle to generalize to new phrasings for the same intent indicating that models cannot understand the semantic content of the given utterance. In this work, we incorporated language models pre-trained on unlabeled text data inside E2E-SLU frameworks to build strong semantic representations. Incorporating both semantic and acoustic information can increase the inference time, leading to high latency when deployed for applications like voice assistants. We developed a 2-pass SLU system that makes low latency prediction using acoustic information from the few seconds of the audio in the first pass and makes higher quality prediction in the second pass by combining semantic and acoustic representations. We take inspiration from prior work on 2-pass end-to-end speech recognition systems that attends on both audio and first-pass hypothesis using a deliberation network. The proposed 2-pass SLU system outperforms the acoustic-based SLU model on the Fluent Speech Commands Challenge Set and SLURP dataset and reduces latency, thus improving user experience. Our code and models are publicly available as part of the ESPnet-SLU toolkit.

Siddhant Arora, Siddharth Dalmia, Brian Yan et al. · cmu

End-to-end spoken language understanding (SLU) systems are gaining popularity over cascaded approaches due to their simplicity and ability to avoid error propagation. However, these systems model sequence labeling as a sequence prediction task causing a divergence from its well-established token-level tagging formulation. We build compositional end-to-end SLU systems that explicitly separate the added complexity of recognizing spoken mentions in SLU from the NLU task of sequence labeling. By relying on intermediate decoders trained for ASR, our end-to-end systems transform the input modality from speech to token-level representations that can be used in the traditional sequence labeling framework. This composition of ASR and NLU formulations in our end-to-end SLU system offers direct compatibility with pre-trained ASR and NLU systems, allows performance monitoring of individual components and enables the use of globally normalized losses like CRF, making them attractive in practical scenarios. Our models outperform both cascaded and direct end-to-end models on a labeling task of named entity recognition across SLU benchmarks.

Roshan Sharma, Kenneth Zheng, Siddhant Arora et al. · cmu, meta-ai

End-to-end speech summarization has been shown to improve performance over cascade baselines. However, such models are difficult to train on very large inputs (dozens of minutes or hours) owing to compute restrictions and are hence trained with truncated model inputs. Truncation leads to poorer models, and a solution to this problem rests in block-wise modeling, i.e., processing a portion of the input frames at a time. In this paper, we develop a method that allows one to train summarization models on very long sequences in an incremental manner. Speech summarization is realized as a streaming process, where hypothesis summaries are updated every block based on new acoustic information. We devise and test strategies to pass semantic context across the blocks. Experiments on the How2 dataset demonstrate that the proposed block-wise training method improves by 3 points absolute on ROUGE-L over a truncated input baseline.

Masao Someki, Kwanghee Choi, Siddhant Arora et al. · cmu, nvidia

We introduce ESPnet-EZ, an extension of the open-source speech processing toolkit ESPnet, aimed at quick and easy development of speech models. ESPnet-EZ focuses on two major aspects: (i) easy fine-tuning and inference of existing ESPnet models on various tasks and (ii) easy integration with popular deep neural network frameworks such as PyTorch-Lightning, Hugging Face transformers and datasets, and Lhotse. By replacing ESPnet design choices inherited from Kaldi with a Python-only, Bash-free interface, we dramatically reduce the effort required to build, debug, and use a new model. For example, to fine-tune a speech foundation model, ESPnet-EZ, compared to ESPnet, reduces the number of newly written code by 2.7x and the amount of dependent code by 6.7x while dramatically reducing the Bash script dependencies. The codebase of ESPnet-EZ is publicly available.

Siddhant Arora, Hayato Futami, Jee-weon Jung et al. · nvidia

Recent studies leverage large language models with multi-tasking capabilities, using natural language prompts to guide the model's behavior and surpassing performance of task-specific models. Motivated by this, we ask: can we build a single model that jointly performs various spoken language understanding (SLU) tasks? We start by adapting a pre-trained automatic speech recognition model to additional tasks using single-token task specifiers. We enhance this approach through instruction tuning, i.e., finetuning by describing the task using natural language instructions followed by the list of label options. Our approach can generalize to new task descriptions for the seen tasks during inference, thereby enhancing its user-friendliness. We demonstrate the efficacy of our single multi-task learning model "UniverSLU" for 12 speech classification and sequence generation task types spanning 17 datasets and 9 languages. On most tasks, UniverSLU achieves competitive performance and often even surpasses task-specific models. Additionally, we assess the zero-shot capabilities, finding that the model generalizes to new datasets and languages for seen task types.

Siddhant Arora, George Saon, Shinji Watanabe et al. · ibm-research

Non-autoregressive (NAR) modeling has gained significant interest in speech processing since these models achieve dramatically lower inference time than autoregressive (AR) models while also achieving good transcription accuracy. Since NAR automatic speech recognition (ASR) models must wait for the completion of the entire utterance before processing, some works explore streaming NAR models based on blockwise attention for low-latency applications. However, streaming NAR models significantly lag in accuracy compared to streaming AR and non-streaming NAR models. To address this, we propose a streaming "semi-autoregressive" ASR model that incorporates the labels emitted in previous blocks as additional context using a Language Model (LM) subnetwork. We also introduce a novel greedy decoding algorithm that addresses insertion and deletion errors near block boundaries while not significantly increasing the inference time. Experiments show that our method outperforms the existing streaming NAR model by 19% relative on Tedlium2, 16%/8% on Librispeech-100 clean/other test sets, and 19%/8% on the Switchboard(SWB)/Callhome(CH) test sets. It also reduced the accuracy gap with streaming AR and non-streaming NAR models while achieving 2.5x lower latency. We also demonstrate that our approach can effectively utilize external text data to pre-train the LM subnetwork to further improve streaming ASR accuracy.

Yosuke Higuchi, Brian Yan, Siddhant Arora et al.

This paper presents BERT-CTC, a novel formulation of end-to-end speech recognition that adapts BERT for connectionist temporal classification (CTC). Our formulation relaxes the conditional independence assumptions used in conventional CTC and incorporates linguistic knowledge through the explicit output dependency obtained by BERT contextual embedding. BERT-CTC attends to the full contexts of the input and hypothesized output sequences via the self-attention mechanism. This mechanism encourages a model to learn inner/inter-dependencies between the audio and token representations while maintaining CTC's training efficiency. During inference, BERT-CTC combines a mask-predict algorithm with CTC decoding, which iteratively refines an output sequence. The experimental results reveal that BERT-CTC improves over conventional approaches across variations in speaking styles and languages. Finally, we show that the semantic representations in BERT-CTC are beneficial towards downstream spoken language understanding tasks.

Zhaojiang Lin, Yong Xu, Kai Sun et al.

Wearable devices such as AI glasses are transforming voice assistants into always-available, hands-free collaborators that integrate seamlessly with daily life, but they also introduce challenges like egocentric audio affected by motion and noise, rapid micro-interactions, and the need to distinguish device-directed speech from background conversations. Existing benchmarks largely overlook these complexities, focusing instead on clean or generic conversational audio. To bridge this gap, we present WearVox, the first benchmark designed to rigorously evaluate voice assistants in realistic wearable scenarios. WearVox comprises 3,842 multi-channel, egocentric audio recordings collected via AI glasses across five diverse tasks including Search-Grounded QA, Closed-Book QA, Side-Talk Rejection, Tool Calling, and Speech Translation, spanning a wide range of indoor and outdoor environments and acoustic conditions. Each recording is accompanied by rich metadata, enabling nuanced analysis of model performance under real-world constraints. We benchmark leading proprietary and open-source speech Large Language Models (SLLMs) and find that most real-time SLLMs achieve accuracies on WearVox ranging from 29% to 59%, with substantial performance degradation on noisy outdoor audio, underscoring the difficulty and realism of the benchmark. Additionally, we conduct a case study with two new SLLMs that perform inference with single-channel and multi-channel audio, demonstrating that multi-channel audio inputs significantly enhance model robustness to environmental noise and improve discrimination between device-directed and background speech. Our results highlight the critical importance of spatial audio cues for context-aware voice assistants and establish WearVox as a comprehensive testbed for advancing wearable voice AI research.

Keqi Deng, Shinji Watanabe, Jiatong Shi et al.

Although Transformers have gained success in several speech processing tasks like spoken language understanding (SLU) and speech translation (ST), achieving online processing while keeping competitive performance is still essential for real-world interaction. In this paper, we take the first step on streaming SLU and simultaneous ST using a blockwise streaming Transformer, which is based on contextual block processing and blockwise synchronous beam search. Furthermore, we design an automatic speech recognition (ASR)-based intermediate loss regularization for the streaming SLU task to improve the classification performance further. As for the simultaneous ST task, we propose a cross-lingual encoding method, which employs a CTC branch optimized with target language translations. In addition, the CTC translation output is also used to refine the search space with CTC prefix score, achieving joint CTC/attention simultaneous translation for the first time. Experiments for SLU are conducted on FSC and SLURP corpora, while the ST task is evaluated on Fisher-CallHome Spanish and MuST-C En-De corpora. Experimental results show that the blockwise streaming Transformer achieves competitive results compared to offline models, especially with our proposed methods that further yield a 2.4% accuracy gain on the SLU task and a 4.3 BLEU gain on the ST task over streaming baselines.

Hayato Futami, Emiru Tsunoo, Kentaro Shibata et al.

Disfluency detection has mainly been solved in a pipeline approach, as post-processing of speech recognition. In this study, we propose Transformer-based encoder-decoder models that jointly solve speech recognition and disfluency detection, which work in a streaming manner. Compared to pipeline approaches, the joint models can leverage acoustic information that makes disfluency detection robust to recognition errors and provide non-verbal clues. Moreover, joint modeling results in low-latency and lightweight inference. We investigate two joint model variants for streaming disfluency detection: a transcript-enriched model and a multi-task model. The transcript-enriched model is trained on text with special tags indicating the starting and ending points of the disfluent part. However, it has problems with latency and standard language model adaptation, which arise from the additional disfluency tags. We propose a multi-task model to solve such problems, which has two output layers at the Transformer decoder; one for speech recognition and the other for disfluency detection. It is modeled to be conditioned on the currently recognized token with an additional token-dependency mechanism. We show that the proposed joint models outperformed a BERT-based pipeline approach in both accuracy and latency, on both the Switchboard and the corpus of spontaneous Japanese.

Sonu Gupta, Ellen Poplavska, Nora O'Toole et al.

The landscape of privacy laws and regulations around the world is complex and ever-changing. National and super-national laws, agreements, decrees, and other government-issued rules form a patchwork that companies must follow to operate internationally. To examine the status and evolution of this patchwork, we introduce the Government Privacy Instructions Corpus, or GPI Corpus, of 1,043 privacy laws, regulations, and guidelines, covering 182 jurisdictions. This corpus enables a large-scale quantitative and qualitative examination of legal foci on privacy. We examine the temporal distribution of when GPIs were created and illustrate the dramatic increase in privacy legislation over the past 50 years, although a finer-grained examination reveals that the rate of increase varies depending on the personal data types that GPIs address. Our exploration also demonstrates that most privacy laws respectively address relatively few personal data types, showing that comprehensive privacy legislation remains rare. Additionally, topic modeling results show the prevalence of common themes in GPIs, such as finance, healthcare, and telecommunications. Finally, we release the corpus to the research community to promote further study.

Siddhant Arora, Jinchuan Tian, Jiatong Shi et al.

Reinforcement learning from human or AI feedback (RLHF/RLAIF) for speech-in/speech-out dialogue systems (SDS) remains underexplored, with prior work largely limited to single semantic rewards applied at the utterance level. Such setups overlook the multi-dimensional and multi-modal nature of conversational quality, which encompasses semantic coherence, audio naturalness, speaker consistency, emotion alignment, and turn-taking behavior. Moreover, they are fundamentally mismatched with duplex spoken dialogue systems that generate responses incrementally, where agents must make decisions based on partial utterances. We address these limitations with the first multi-reward RLAIF framework for SDS, combining semantic, audio-quality, and emotion-consistency rewards. To align utterance-level preferences with incremental, blockwise decoding in duplex models, we apply turn-level preference sampling and aggregate per-block log-probabilities within a single DPO objective. We present the first systematic study of preference learning for improving SDS quality in both multi-turn Chain-of-Thought and blockwise duplex models, and release a multi-reward DPO dataset to support reproducible research. Experiments show that single-reward RLAIF selectively improves its targeted metric, while joint multi-reward training yields consistent gains across semantic quality and audio naturalness. These results highlight the importance of holistic, multi-reward alignment for practical conversational SDS.

Yao-Fei Cheng, Hayato Futami, Yosuke Kashiwagi et al.

Recent progress in large language models (LLMs) has gained interest in speech-text multimodal foundation models, achieving strong performance on instruction-tuned speech translation (ST). However, expanding language pairs is costly due to re-training on combined new and previous datasets. To address this, we aim to build a one-to-many ST system from existing one-to-one ST systems using task arithmetic without re-training. Direct application of task arithmetic in ST leads to language confusion; therefore, we introduce an augmented task arithmetic method incorporating a language control model to ensure correct target language generation. Our experiments on MuST-C and CoVoST-2 show BLEU score improvements of up to 4.66 and 4.92, with COMET gains of 8.87 and 11.83. In addition, we demonstrate our framework can extend to language pairs lacking paired ST training data or pre-trained ST models by synthesizing ST models based on existing machine translation (MT) and ST models via task analogies.

Siddhant Arora, Hayato Futami, Yosuke Kashiwagi et al.

There has been an increased interest in the integration of pretrained speech recognition (ASR) and language models (LM) into the SLU framework. However, prior methods often struggle with a vocabulary mismatch between pretrained models, and LM cannot be directly utilized as they diverge from its NLU formulation. In this study, we propose a three-pass end-to-end (E2E) SLU system that effectively integrates ASR and LM subnetworks into the SLU formulation for sequence generation tasks. In the first pass, our architecture predicts ASR transcripts using the ASR subnetwork. This is followed by the LM subnetwork, which makes an initial SLU prediction. Finally, in the third pass, the deliberation subnetwork conditions on representations from the ASR and LM subnetworks to make the final prediction. Our proposed three-pass SLU system shows improved performance over cascaded and E2E SLU models on two benchmark SLU datasets, SLURP and SLUE, especially on acoustically challenging utterances.

Yifan Peng, Jinchuan Tian, William Chen et al. · nvidia

Recent studies have highlighted the importance of fully open foundation models. The Open Whisper-style Speech Model (OWSM) is an initial step towards reproducing OpenAI Whisper using public data and open-source toolkits. However, previous versions of OWSM (v1 to v3) are still based on standard Transformer, which might lead to inferior performance compared to state-of-the-art speech encoder architectures. This work aims to improve the performance and efficiency of OWSM without additional data. We present a series of E-Branchformer-based models named OWSM v3.1, ranging from 100M to 1B parameters. OWSM v3.1 outperforms its predecessor, OWSM v3, in most evaluation benchmarks, while showing an improved inference speed of up to 25%. We further reveal the emergent ability of OWSM v3.1 in zero-shot contextual biasing speech recognition. We also provide a model trained on a subset of data with low license restrictions. We will publicly release the code, pre-trained models, and training logs.

Chien-yu Huang, Wei-Chih Chen, Shu-wen Yang et al. · cmu, mit

Multimodal foundation models, such as Gemini and ChatGPT, have revolutionized human-machine interactions by seamlessly integrating various forms of data. Developing a universal spoken language model that comprehends a wide range of natural language instructions is critical for bridging communication gaps and facilitating more intuitive interactions. However, the absence of a comprehensive evaluation benchmark poses a significant challenge. We present Dynamic-SUPERB Phase-2, an open and evolving benchmark for the comprehensive evaluation of instruction-based universal speech models. Building upon the first generation, this second version incorporates 125 new tasks contributed collaboratively by the global research community, expanding the benchmark to a total of 180 tasks, making it the largest benchmark for speech and audio evaluation. While the first generation of Dynamic-SUPERB was limited to classification tasks, Dynamic-SUPERB Phase-2 broadens its evaluation capabilities by introducing a wide array of novel and diverse tasks, including regression and sequence generation, across speech, music, and environmental audio. Evaluation results show that no model performed well universally. SALMONN-13B excelled in English ASR and Qwen2-Audio-7B-Instruct showed high accuracy in emotion recognition, but current models still require further innovations to handle a broader range of tasks. We open-source all task data and the evaluation pipeline at https://github.com/dynamic-superb/dynamic-superb.

Yosuke Kashiwagi, Hayato Futami, Emiru Tsunoo et al.

In many real-world scenarios, such as meetings, multiple speakers are present with an unknown number of participants, and their utterances often overlap. We address these multi-speaker challenges by a novel attention-based encoder-decoder method augmented with special speaker class tokens obtained by speaker clustering. During inference, we select multiple recognition hypotheses conditioned on predicted speaker cluster tokens, and these hypotheses are merged by agglomerative hierarchical clustering (AHC) based on the normalized edit distance. The clustered hypotheses result in the multi-speaker transcriptions with the appropriate number of speakers determined by AHC. Our experiments on the LibriMix dataset demonstrate that our proposed method was particularly effective in complex 3-mix environments, achieving a 55% relative error reduction on clean data and a 36% relative error reduction on noisy data compared with conventional serialized output training.

Jinchuan Tian, Haoran Wang, Bo-Hao Su et al.

Current audio foundation models typically rely on rigid, task-specific supervision, addressing isolated factors of audio rather than the whole. In contrast, human intelligence processes audio holistically, seamlessly bridging physical signals with abstract cognitive concepts to execute complex tasks. Grounded in this philosophy, we introduce Bagpiper, an 8B audio foundation model that interprets physical audio via rich captions, i.e., comprehensive natural language descriptions that encapsulate the critical cognitive concepts inherent in the signal (e.g., transcription, audio events). By pre-training on a massive corpus of 600B tokens, the model establishes a robust bidirectional mapping between raw audio and this high-level conceptual space. During fine-tuning, Bagpiper adopts a caption-then-process workflow, simulating an intermediate cognitive reasoning step to solve diverse tasks without task-specific priors. Experimentally, Bagpiper outperforms Qwen-2.5-Omni on MMAU and AIRBench for audio understanding and surpasses CosyVoice3 and TangoFlux in generation quality, capable of synthesizing arbitrary compositions of speech, music, and sound effects. To the best of our knowledge, Bagpiper is among the first works that achieve unified understanding generation for general audio. Model, data, and code are available at Bagpiper Home Page.

Siddhant Arora, Zhiyun Lu, Chung-Cheng Chiu et al.

The recent wave of audio foundation models (FMs) could provide new capabilities for conversational modeling. However, there have been limited efforts to evaluate these audio FMs comprehensively on their ability to have natural and interactive conversations. To engage in meaningful conversation with the end user, we would want the FMs to additionally perform a fluent succession of turns without too much overlapping speech or long stretches of silence. Inspired by this, we ask whether the recently proposed audio FMs can understand, predict, and perform turn-taking events? To answer this, we propose a novel evaluation protocol that can assess spoken dialog system's turn-taking capabilities using a supervised model as a judge that has been trained to predict turn-taking events in human-human conversations. Using this protocol, we present the first comprehensive user study that evaluates existing spoken dialogue systems on their ability to perform turn-taking events and reveal many interesting insights, such as they sometimes do not understand when to speak up, can interrupt too aggressively and rarely backchannel. We further evaluate multiple open-source and proprietary audio FMs accessible through APIs on carefully curated test benchmarks from Switchboard to measure their ability to understand and predict turn-taking events and identify significant room for improvement. We will open source our evaluation platform to promote the development of advanced conversational AI systems.

Jinchuan Tian, Jiatong Shi, William Chen et al. · nvidia

We present ESPnet-SpeechLM, an open toolkit designed to democratize the development of speech language models (SpeechLMs) and voice-driven agentic applications. The toolkit standardizes speech processing tasks by framing them as universal sequential modeling problems, encompassing a cohesive workflow of data preprocessing, pre-training, inference, and task evaluation. With ESPnet-SpeechLM, users can easily define task templates and configure key settings, enabling seamless and streamlined SpeechLM development. The toolkit ensures flexibility, efficiency, and scalability by offering highly configurable modules for every stage of the workflow. To illustrate its capabilities, we provide multiple use cases demonstrating how competitive SpeechLMs can be constructed with ESPnet-SpeechLM, including a 1.7B-parameter model pre-trained on both text and speech tasks, across diverse benchmarks. The toolkit and its recipes are fully transparent and reproducible at: https://github.com/espnet/espnet/tree/speechlm.

Jinchuan Tian, William Chen, Yifan Peng et al. · nvidia

This paper presents Open Unified Speech Language Models (OpusLMs), a family of open foundational speech language models (SpeechLMs) up to 7B. Initialized from decoder-only text language models, the OpusLMs are continuously pre-trained on 213K hours of speech-text pairs and 292B text-only tokens. We demonstrate our OpusLMs achieve comparable (or even superior) performance with existing SpeechLMs in speech recognition, speech synthesis, and text-only capabilities. Technically, this paper articulates our SpeechLM designs on tokenization, multi-stream language models, and multi-stage training strategies. We experimentally demonstrate the importance of model size scaling and the effect of annealing data selection. The OpusLMs are all built from publicly available materials and are fully transparent models. We release our code, data, checkpoints, and training logs to facilitate open SpeechLM research

Siddhant Arora, Yifan Peng, Jiatong Shi et al. · nvidia

Advancements in audio foundation models (FMs) have fueled interest in end-to-end (E2E) spoken dialogue systems, but different web interfaces for each system makes it challenging to compare and contrast them effectively. Motivated by this, we introduce an open-source, user-friendly toolkit designed to build unified web interfaces for various cascaded and E2E spoken dialogue systems. Our demo further provides users with the option to get on-the-fly automated evaluation metrics such as (1) latency, (2) ability to understand user input, (3) coherence, diversity, and relevance of system response, and (4) intelligibility and audio quality of system output. Using the evaluation metrics, we compare various cascaded and E2E spoken dialogue systems with a human-human conversation dataset as a proxy. Our analysis demonstrates that the toolkit allows researchers to effortlessly compare and contrast different technologies, providing valuable insights such as current E2E systems having poorer audio quality and less diverse responses. An example demo produced using our toolkit is publicly available here: https://huggingface.co/spaces/Siddhant/Voice_Assistant_Demo.

Orevaoghene Ahia, Martijn Bartelds, Kabir Ahuja et al.

Developing large audio language models (LMs) capable of understanding diverse spoken interactions is essential for accommodating the multimodal nature of human communication and can increase the accessibility of language technologies across different user populations. Recent work on audio LMs has primarily evaluated their performance on short audio segments, typically under 30 seconds, with limited exploration of long-form conversational speech segments that more closely reflect natural user interactions with these models. We introduce Brutally Long Audio Bench (BLAB), a challenging long-form audio benchmark that evaluates audio LMs on localization, duration estimation, emotion, and counting tasks using audio segments averaging 51 minutes in length. BLAB consists of 833+ hours of diverse, full-length audio clips, each paired with human-annotated, text-based natural language questions and answers. Our audio data were collected from permissively licensed sources and underwent a human-assisted filtering process to ensure task compliance. We evaluate six open-source and proprietary audio LMs on BLAB and find that all of them, including advanced models such as Gemini 2.0 Pro and GPT-4o, struggle with the tasks in BLAB. Our comprehensive analysis reveals key insights into the trade-offs between task difficulty and audio duration. In general, we find that audio LMs struggle with long-form speech, with performance declining as duration increases. They perform poorly on localization, temporal reasoning, counting, and struggle to understand non-phonemic information, relying more on prompts than audio content. BLAB serves as a challenging evaluation framework to develop audio LMs with robust long-form audio understanding capabilities.

Leander Melroy Maben, Gayathri Ganesh Lakshmy, Srijith Radhakrishnan et al.

Despite advances in language and speech technologies, no open-source system enables full speech-to-speech, multi-turn dialogue with integrated tool use and agentic reasoning. We introduce AURA (Agent for Understanding, Reasoning, and Automated Tool Use), the first open-source, speech-native assistant capable of completing complex, goal-driven tasks through dynamic tool invocation and multi-turn conversation. AURA combines open-weight ASR, TTS, and LLMs in a cascaded pipeline and supports tools such as calendar booking, contact lookup, web search, and email. Its modular design allows easy integration of new tools using natural language prompts and action classes. On VoiceBench, AURA scores 92.75% on OpenBookQA-outperforming all open-weight systems and nearing GPT-4o-and 4.39 on AlpacaEval, competitive with other open-weight systems. Human evaluation shows 90% task success on complex, multi-turn speech tasks.

Siddhant Arora, Ankita Pasad, Chung-Ming Chien et al.

The Spoken Language Understanding Evaluation (SLUE) suite of benchmark tasks was recently introduced to address the need for open resources and benchmarking of complex spoken language understanding (SLU) tasks, including both classification and sequence generation tasks, on natural speech. The benchmark has demonstrated preliminary success in using pre-trained speech foundation models (SFM) for these SLU tasks. However, the community still lacks a fine-grained understanding of the comparative utility of different SFMs. Inspired by this, we ask: which SFMs offer the most benefits for these complex SLU tasks, and what is the most effective approach for incorporating these SFMs? To answer this, we perform an extensive evaluation of multiple supervised and self-supervised SFMs using several evaluation protocols: (i) frozen SFMs with a lightweight prediction head, (ii) frozen SFMs with a complex prediction head, and (iii) fine-tuned SFMs with a lightweight prediction head. Although the supervised SFMs are pre-trained on much more speech recognition data (with labels), they do not always outperform self-supervised SFMs; the latter tend to perform at least as well as, and sometimes better than, supervised SFMs, especially on the sequence generation tasks in SLUE. While there is no universally optimal way of incorporating SFMs, the complex prediction head gives the best performance for most tasks, although it increases the inference time. We also introduce an open-source toolkit and performance leaderboard, SLUE-PERB, for these tasks and modeling strategies.

Siddhant Arora, Siddharth Dalmia, Pavel Denisov et al.

As Automatic Speech Processing (ASR) systems are getting better, there is an increasing interest of using the ASR output to do downstream Natural Language Processing (NLP) tasks. However, there are few open source toolkits that can be used to generate reproducible results on different Spoken Language Understanding (SLU) benchmarks. Hence, there is a need to build an open source standard that can be used to have a faster start into SLU research. We present ESPnet-SLU, which is designed for quick development of spoken language understanding in a single framework. ESPnet-SLU is a project inside end-to-end speech processing toolkit, ESPnet, which is a widely used open-source standard for various speech processing tasks like ASR, Text to Speech (TTS) and Speech Translation (ST). We enhance the toolkit to provide implementations for various SLU benchmarks that enable researchers to seamlessly mix-and-match different ASR and NLU models. We also provide pretrained models with intensively tuned hyper-parameters that can match or even outperform the current state-of-the-art performances. The toolkit is publicly available at https://github.com/espnet/espnet.

Siddhant Arora, Kai-Wei Chang, Chung-Ming Chien et al. · mit, nvidia

The field of spoken language processing is undergoing a shift from training custom-built, task-specific models toward using and optimizing spoken language models (SLMs) which act as universal speech processing systems. This trend is similar to the progression toward universal language models that has taken place in the field of (text) natural language processing. SLMs include both "pure" language models of speech -- models of the distribution of tokenized speech sequences -- and models that combine speech encoders with text language models, often including both spoken and written input or output. Work in this area is very diverse, with a range of terminology and evaluation settings. This paper aims to contribute an improved understanding of SLMs via a unifying literature survey of recent work in the context of the evolution of the field. Our survey categorizes the work in this area by model architecture, training, and evaluation choices, and describes some key challenges and directions for future work.

Hayato Futami, Emiru Tsunoo, Yosuke Kashiwagi et al.

In speech recognition applications, it is important to recognize context-specific rare words, such as proper nouns. Tree-constrained Pointer Generator (TCPGen) has shown promise for this purpose, which efficiently biases such words with a prefix tree. While the original TCPGen relies on grapheme-based encoding, we propose extending it with phoneme-aware encoding to better recognize words of unusual pronunciations. As TCPGen handles biasing words as subword units, we propose obtaining subword-level phoneme-aware encoding by using alignment between phonemes and subwords. Furthermore, we propose injecting phoneme-level predictions from CTC into queries of TCPGen so that the model better interprets the phoneme-aware encodings. We conducted ASR experiments with TCPGen for RNN transducer. We observed that proposed phoneme-aware encoding outperformed ordinary grapheme-based encoding on both the English LibriSpeech and Japanese CSJ datasets, demonstrating the robustness of our approach across linguistically diverse languages.

Minsu Kim, Jee-weon Jung, Hyeongseop Rha et al.

The capability to jointly process multi-modal information is becoming an essential task. However, the limited number of paired multi-modal data and the large computational requirements in multi-modal learning hinder the development. We propose a novel Tri-Modal Translation (TMT) model that translates between arbitrary modalities spanning speech, image, and text. We introduce a novel viewpoint, where we interpret different modalities as different languages, and treat multi-modal translation as a well-established machine translation problem. To this end, we tokenize speech and image data into discrete tokens, which provide a unified interface across modalities and significantly decrease the computational cost. In the proposed TMT, a multi-modal encoder-decoder conducts the core translation, whereas modality-specific processing is conducted only within the tokenization and detokenization stages. We evaluate the proposed TMT on all six modality translation tasks. TMT outperforms single model counterparts consistently, demonstrating that unifying tasks is beneficial not only for practicality but also for performance.

Siddhant Arora, Jinchuan Tian, Hayato Futami et al.

Unlike traditional cascaded pipelines, end-to-end (E2E) spoken dialogue systems preserve full differentiability and capture non-phonemic information, making them well-suited for modeling spoken interactions. However, existing E2E approaches often require large-scale training data and generates responses lacking semantic coherence. We propose a simple yet effective strategy leveraging a chain-of-thought (CoT) formulation, ensuring that training on conversational data remains closely aligned with the multimodal language model (LM)'s pre-training on speech recognition~(ASR), text-to-speech synthesis (TTS), and text LM tasks. Our method achieves over 1.5 ROUGE-1 improvement over the baseline, successfully training spoken dialogue systems on publicly available human-human conversation datasets, while being compute-efficient enough to train on just 300 hours of public human-human conversation data, such as the Switchboard. We will publicly release our models and training code.

Siddhant Arora, Jinchuan Tian, Hayato Futami et al.

Most end-to-end (E2E) spoken dialogue systems (SDS) rely on voice activity detection (VAD) for turn-taking, but VAD fails to distinguish between pauses and turn completions. Duplex SDS models address this by predicting output continuously, including silence tokens, thus removing the need for explicit VAD. However, they often have complex dual-channel architecture and lag behind cascaded models in semantic reasoning. To overcome these challenges, we propose SCoT: a Streaming Chain-of-Thought (CoT) framework for Duplex SDS, alternating between processing fixed-duration user input and generating responses in a blockwise manner. Using frame-level alignments, we create intermediate targets-aligned user transcripts and system responses for each block. Experiments show that our approach produces more coherent and interpretable responses than existing duplex methods while supporting lower-latency and overlapping interactions compared to turn-by-turn systems.

Siddhant Arora, Haidar Khan, Kai Sun et al.

End-to-end speech-in speech-out dialogue systems are emerging as a powerful alternative to traditional ASR-LLM-TTS pipelines, generating more natural, expressive responses with significantly lower latency. However, these systems remain prone to hallucinations due to limited factual grounding. While text-based dialogue systems address this challenge by integrating tools such as web search and knowledge graph APIs, we introduce the first approach to extend tool use directly into speech-in speech-out systems. A key challenge is that tool integration substantially increases response latency, disrupting conversational flow. To mitigate this, we propose Streaming Retrieval-Augmented Generation (Streaming RAG), a novel framework that reduces user-perceived latency by predicting tool queries in parallel with user speech, even before the user finishes speaking. Specifically, we develop a post-training pipeline that teaches the model when to issue tool calls during ongoing speech and how to generate spoken summaries that fuse audio queries with retrieved text results, thereby improving both accuracy and responsiveness. To evaluate our approach, we construct AudioCRAG, a benchmark created by converting queries from the publicly available CRAG dataset into speech form. Experimental results demonstrate that our streaming RAG approach increases QA accuracy by up to 200% relative (from 11.1% to 34.2% absolute) and further enhances user experience by reducing tool use latency by 20%. Importantly, our streaming RAG approach is modality-agnostic and can be applied equally to typed input, paving the way for more agentic, real-time AI assistants.

Emiru Tsunoo, Hayato Futami, Yosuke Kashiwagi et al.

For streaming speech recognition, a Transformer-based encoder has been widely used with block processing. Although many studies addressed improving emission latency of transducers, little work has been explored for improving encoding latency of the block processing. We seek to reduce latency by frequently emitting a chunk with a small shift rather than scarce large-chunk emissions, resulting in higher computational costs. To efficiently compute with the small chunk shift, we propose a new encoder, Spiralformer, tailored for block processing by combining layer dropping and early exiting. We skip layer computation in a cyclic manner and shift the computed layer in each block spirally, which completes computation for all the layers over the block processing. Experimentally, we observed that our method achieved 21.6% reduction in the averaged token emission delay in Librispeech, and 7.0% in CSJ, compared with the baseline with similar computational cost and word error rates.

Anyu Ying, Natarajan Balaji Shankar, Chyi-Jiunn Lin et al.

Despite advancements in ASR, child speech recognition remains challenging due to acoustic variability and limited annotated data. While fine-tuning adult ASR models on child speech is common, comparisons with flat-start training remain underexplored. We compare flat-start training across multiple datasets, SSL representations (WavLM, XEUS), and decoder architectures. Our results show that SSL representations are biased toward adult speech, with flat-start training on child speech mitigating these biases. We also analyze model scaling, finding consistent improvements up to 1B parameters, beyond which performance plateaus. Additionally, age-related ASR and speaker verification analysis highlights the limitations of proprietary models like Whisper, emphasizing the need for open-data models for reliable child speech research. All investigations are conducted using ESPnet, and our publicly available benchmark provides insights into training strategies for robust child speech processing.

Hayato Futami, Emiru Tsunoo, Yosuke Kashiwagi et al.

Speech-to-speech translation (S2ST) has been advanced with large language models (LLMs), which are fine-tuned on discrete speech units. In such approaches, modality adaptation from text to speech has been an issue. LLMs are trained on text-only data, which presents challenges to adapt them to speech modality with limited speech-to-speech data. To address the training difficulty, we propose scheduled interleaved speech--text training in this study. We use interleaved speech--text units instead of speech units during training, where aligned text tokens are interleaved at the word level. We gradually decrease the ratio of text as training progresses, to facilitate progressive modality adaptation from text to speech. We conduct experimental evaluations by fine-tuning LLaMA3.2-1B for S2ST on the CVSS dataset. We show that the proposed method consistently improves the translation performances, especially for languages with limited training data.

Emiru Tsunoo, Hayato Futami, Yosuke Kashiwagi et al.

Decoder-only language models (LMs) have been successfully adopted for speech-processing tasks including automatic speech recognition (ASR). The LMs have ample expressiveness and perform efficiently. This efficiency is a suitable characteristic for streaming applications of ASR. In this work, we propose to use a decoder-only architecture for blockwise streaming ASR. In our approach, speech features are compressed using CTC output and context embedding using blockwise speech subnetwork, and are sequentially provided as prompts to the decoder. The decoder estimates the output tokens promptly at each block. To this end, we also propose a novel training scheme using random-length prefix prompts to make the model robust to the truncated prompts caused by blockwise processing. An experimental comparison shows that our proposed decoder-only streaming ASR achieves 8% relative word error rate reduction in the LibriSpeech test-other set while being twice as fast as the baseline model.

Yosuke Kashiwagi, Hayato Futami, Emiru Tsunoo et al.

End-to-end multilingual speech recognition models handle multiple languages through a single model, often incorporating language identification to automatically detect the language of incoming speech. Since the common scenario is where the language is already known, these models can perform as language-specific by using language information as prompts, which is particularly beneficial for attention-based encoder-decoder architectures. However, the Connectionist Temporal Classification (CTC) approach, which enhances recognition via joint decoding and multi-task training, does not normally incorporate language prompts due to its conditionally independent output tokens. To overcome this, we introduce an encoder prompting technique within the self-conditioned CTC framework, enabling language-specific adaptation of the CTC model in a zero-shot manner. Our method has shown to significantly reduce errors by 28% on average and by 41% on low-resource languages.

Hayato Futami, Siddhant Arora, Yosuke Kashiwagi et al.

Recently, multi-task spoken language understanding (SLU) models have emerged, designed to address various speech processing tasks. However, these models often rely on a large number of parameters. Also, they often encounter difficulties in adapting to new data for a specific task without experiencing catastrophic forgetting of previously trained tasks. In this study, we propose finding task-specific subnetworks within a multi-task SLU model via neural network pruning. In addition to model compression, we expect that the forgetting of previously trained tasks can be mitigated by updating only a task-specific subnetwork. We conduct experiments on top of the state-of-the-art multi-task SLU model ``UniverSLU'', trained for several tasks such as emotion recognition (ER), intent classification (IC), and automatic speech recognition (ASR). We show that pruned models were successful in adapting to additional ASR or IC data with minimal performance degradation on previously trained tasks.

Yifan Peng, Kwangyoun Kim, Felix Wu et al.

Conformer, a convolution-augmented Transformer variant, has become the de facto encoder architecture for speech processing due to its superior performance in various tasks, including automatic speech recognition (ASR), speech translation (ST) and spoken language understanding (SLU). Recently, a new encoder called E-Branchformer has outperformed Conformer in the LibriSpeech ASR benchmark, making it promising for more general speech applications. This work compares E-Branchformer and Conformer through extensive experiments using different types of end-to-end sequence-to-sequence models. Results demonstrate that E-Branchformer achieves comparable or better performance than Conformer in almost all evaluation sets across 15 ASR, 2 ST, and 3 SLU benchmarks, while being more stable during training. We will release our training configurations and pre-trained models for reproducibility, which can benefit the speech community.

Siddhant Arora, Hayato Futami, Shih-Lun Wu et al.

Recently there have been efforts to introduce new benchmark tasks for spoken language understanding (SLU), like semantic parsing. In this paper, we describe our proposed spoken semantic parsing system for the quality track (Track 1) in Spoken Language Understanding Grand Challenge which is part of ICASSP Signal Processing Grand Challenge 2023. We experiment with both end-to-end and pipeline systems for this task. Strong automatic speech recognition (ASR) models like Whisper and pretrained Language models (LM) like BART are utilized inside our SLU framework to boost performance. We also investigate the output level combination of various models to get an exact match accuracy of 80.8, which won the 1st place at the challenge.

Hayato Futami, Jessica Huynh, Siddhant Arora et al.

This paper describes our system for the low-resource domain adaptation track (Track 3) in Spoken Language Understanding Grand Challenge, which is a part of ICASSP Signal Processing Grand Challenge 2023. In the track, we adopt a pipeline approach of ASR and NLU. For ASR, we fine-tune Whisper for each domain with upsampling. For NLU, we fine-tune BART on all the Track3 data and then on low-resource domain data. We apply masked LM (MLM) -based data augmentation, where some of input tokens and corresponding target labels are replaced using MLM. We also apply a retrieval-based approach, where model input is augmented with similar training samples. As a result, we achieved exact match (EM) accuracy 63.3/75.0 (average: 69.15) for reminder/weather domain, and won the 1st place at the challenge.

Siddhant Arora, Hayato Futami, Emiru Tsunoo et al.

Most human interactions occur in the form of spoken conversations where the semantic meaning of a given utterance depends on the context. Each utterance in spoken conversation can be represented by many semantic and speaker attributes, and there has been an interest in building Spoken Language Understanding (SLU) systems for automatically predicting these attributes. Recent work has shown that incorporating dialogue history can help advance SLU performance. However, separate models are used for each SLU task, leading to an increase in inference time and computation cost. Motivated by this, we aim to ask: can we jointly model all the SLU tasks while incorporating context to facilitate low-latency and lightweight inference? To answer this, we propose a novel model architecture that learns dialog context to jointly predict the intent, dialog act, speaker role, and emotion for the spoken utterance. Note that our joint prediction is based on an autoregressive model and we need to decide the prediction order of dialog attributes, which is not trivial. To mitigate the issue, we also propose an order agnostic training method. Our experiments show that our joint model achieves similar results to task-specific classifiers and can effectively integrate dialog context to further improve the SLU performance.

Siddhant Arora, Danish Pruthi, Norman Sadeh et al.

In attempts to "explain" predictions of machine learning models, researchers have proposed hundreds of techniques for attributing predictions to features that are deemed important. While these attributions are often claimed to hold the potential to improve human "understanding" of the models, surprisingly little work explicitly evaluates progress towards this aspiration. In this paper, we conduct a crowdsourcing study, where participants interact with deception detection models that have been trained to distinguish between genuine and fake hotel reviews. They are challenged both to simulate the model on fresh reviews, and to edit reviews with the goal of lowering the probability of the originally predicted class. Successful manipulations would lead to an adversarial example. During the training (but not the test) phase, input spans are highlighted to communicate salience. Through our evaluation, we observe that for a linear bag-of-words model, participants with access to the feature coefficients during training are able to cause a larger reduction in model confidence in the testing phase when compared to the no-explanation control. For the BERT-based classifier, popular local explanations do not improve their ability to reduce the model confidence over the no-explanation case. Remarkably, when the explanation for the BERT model is given by the (global) attributions of a linear model trained to imitate the BERT model, people can effectively manipulate the model.

Siddhant Arora, Alissa Ostapenko, Vijay Viswanathan et al.

Decomposable tasks are complex and comprise of a hierarchy of sub-tasks. Spoken intent prediction, for example, combines automatic speech recognition and natural language understanding. Existing benchmarks, however, typically hold out examples for only the surface-level sub-task. As a result, models with similar performance on these benchmarks may have unobserved performance differences on the other sub-tasks. To allow insightful comparisons between competitive end-to-end architectures, we propose a framework to construct robust test sets using coordinate ascent over sub-task specific utility functions. Given a dataset for a decomposable task, our method optimally creates a test set for each sub-task to individually assess sub-components of the end-to-end model. Using spoken language understanding as a case study, we generate new splits for the Fluent Speech Commands and Snips SmartLights datasets. Each split has two test sets: one with held-out utterances assessing natural language understanding abilities, and one with held-out speakers to test speech processing skills. Our splits identify performance gaps up to 10% between end-to-end systems that were within 1% of each other on the original test sets. These performance gaps allow more realistic and actionable comparisons between different architectures, driving future model development. We release our splits and tools for the community.

Siddhant Arora, Vinayak Gupta, Garima Gaur et al.

In this paper, we address the problem of learning low dimension representation of entities on relational databases consisting of multiple tables. Embeddings help to capture semantics encoded in the database and can be used in a variety of settings like auto-completion of tables, fully-neural query processing of relational joins queries, seamlessly handling missing values, and more. Current work is restricted to working with just single table, or using pretrained embeddings over an external corpus making them unsuitable for use in real-world databases. In this work, we look into ways of using these attention-based model to learn embeddings for entities in the relational database. We are inspired by BERT style pretraining methods and are interested in observing how they can be extended for representation learning on structured databases. We evaluate our approach of the autocompletion of relational databases and achieve improvement over standard baselines.

Siddhant Arora

Knowledge Graphs are increasingly becoming popular for a variety of downstream tasks like Question Answering and Information Retrieval. However, the Knowledge Graphs are often incomplete, thus leading to poor performance. As a result, there has been a lot of interest in the task of Knowledge Base Completion. More recently, Graph Neural Networks have been used to capture structural information inherently stored in these Knowledge Graphs and have been shown to achieve SOTA performance across a variety of datasets. In this survey, we understand the various strengths and weaknesses of the proposed methodology and try to find new exciting research problems in this area that require further investigation.