WHISMA: A Speech-LLM to Perform Zero-shot Spoken Language Understanding

Speech large language models (speech-LLMs) integrate speech and text-based foundation models to provide a unified framework for handling a wide range of downstream tasks. In this paper, we introduce WHISMA, a speech-LLM tailored for spoken language understanding (SLU) that demonstrates robust performance in various zero-shot settings. WHISMA combines the speech encoder from Whisper with the Llama-3 LLM, and is fine-tuned in a parameter-efficient manner on a comprehensive collection of SLU-related datasets. Our experiments show that WHISMA significantly improves the zero-shot slot filling performance on the SLURP benchmark, achieving a relative gain of 26.6% compared to the current state-of-the-art model. Furthermore, to evaluate WHISMA's generalisation capabilities to unseen domains, we develop a new task-agnostic benchmark named SLU-GLUE. The evaluation results indicate that WHISMA outperforms an existing speech-LLM (Qwen-Audio) with a relative gain of 33.0%.

Prompting Whisper for QA-driven Zero-shot End-to-end Spoken Language Understanding

Zero-shot spoken language understanding (SLU) enables systems to comprehend user utterances in new domains without prior exposure to training data. Recent studies often rely on large language models (LLMs), leading to excessive footprints and complexity. This paper proposes the use of Whisper, a standalone speech processing model, for zero-shot end-to-end (E2E) SLU. To handle unseen semantic labels, SLU tasks are integrated into a question-answering (QA) framework, which prompts the Whisper decoder for semantics deduction. The system is efficiently trained with prefix-tuning, optimising a minimal set of parameters rather than the entire Whisper model. We show that the proposed system achieves a 40.7% absolute gain for slot filling (SLU-F1) on SLURP compared to a recently introduced zero-shot benchmark. Furthermore, it performs comparably to a Whisper-GPT-2 modular system under both in-corpus and cross-corpus evaluation settings, but with a relative 34.8% reduction in model parameters.

DiaLoc: An Iterative Approach to Embodied Dialog Localization

Multimodal learning has advanced the performance for many vision-language tasks. However, most existing works in embodied dialog research focus on navigation and leave the localization task understudied. The few existing dialog-based localization approaches assume the availability of entire dialog prior to localizaiton, which is impractical for deployed dialog-based localization. In this paper, we propose DiaLoc, a new dialog-based localization framework which aligns with a real human operator behavior. Specifically, we produce an iterative refinement of location predictions which can visualize current pose believes after each dialog turn. DiaLoc effectively utilizes the multimodal data for multi-shot localization, where a fusion encoder fuses vision and dialog information iteratively. We achieve state-of-the-art results on embodied dialog-based localization task, in single-shot (+7.08% in Acc5@valUnseen) and multi-shot settings (+10.85% in Acc5@valUnseen). DiaLoc narrows the gap between simulation and real-world applications, opening doors for future research on collaborative localization and navigation.

Self-regularised Minimum Latency Training for Streaming Transformer-based Speech Recognition

This paper proposes a self-regularised minimum latency training (SR-MLT) method for streaming Transformer-based automatic speech recognition (ASR) systems. In previous works, latency was optimised by truncating the online attention weights based on the hard alignments obtained from conventional ASR models, without taking into account the potential loss of ASR accuracy. On the contrary, here we present a strategy to obtain the alignments as a part of the model training without external supervision. The alignments produced by the proposed method are dynamically regularised on the training data, such that the latency reduction does not result in the loss of ASR accuracy. SR-MLT is applied as a fine-tuning step on the pre-trained Transformer models that are based on either monotonic chunkwise attention (MoChA) or cumulative attention (CA) algorithms for online decoding. ASR experiments on the AIShell-1 and Librispeech datasets show that when applied on a decent pre-trained MoChA or CA baseline model, SR-MLT can effectively reduce the latency with the relative gains ranging from 11.8% to 39.5%. Furthermore, we also demonstrate that under certain accuracy levels, the models trained with SR-MLT can achieve lower latency when compared to those supervised using external hard alignments.

Non-autoregressive End-to-end Approaches for Joint Automatic Speech Recognition and Spoken Language Understanding

This paper presents the use of non-autoregressive (NAR) approaches for joint automatic speech recognition (ASR) and spoken language understanding (SLU) tasks. The proposed NAR systems employ a Conformer encoder that applies connectionist temporal classification (CTC) to transcribe the speech utterance into raw ASR hypotheses, which are further refined with a bidirectional encoder representations from Transformers (BERT)-like decoder. In the meantime, the intent and slot labels of the utterance are predicted simultaneously using the same decoder. Both Mask-CTC and self-conditioned CTC (SC-CTC) approaches are explored for this study. Experiments conducted on the SLURP dataset show that the proposed SC-Mask-CTC NAR system achieves 3.7% and 3.2% absolute gains in SLU metrics and a competitive level of ASR accuracy, when compared to a Conformer-Transformer based autoregressive (AR) model. Additionally, the NAR systems achieve 6x faster decoding speed than the AR baseline.

Multiple-hypothesis RNN-T Loss for Unsupervised Fine-tuning and Self-training of Neural Transducer

This paper proposes a new approach to perform unsupervised fine-tuning and self-training using unlabeled speech data for recurrent neural network (RNN)-Transducer (RNN-T) end-to-end (E2E) automatic speech recognition (ASR) systems. Conventional systems perform fine-tuning/self-training using ASR hypothesis as the targets when using unlabeled audio data and are susceptible to the ASR performance of the base model. Here in order to alleviate the influence of ASR errors while using unlabeled data, we propose a multiple-hypothesis RNN-T loss that incorporates multiple ASR 1-best hypotheses into the loss function. For the fine-tuning task, ASR experiments on Librispeech show that the multiple-hypothesis approach achieves a relative reduction of 14.2% word error rate (WER) when compared to the single-hypothesis approach, on the test_other set. For the self-training task, ASR models are trained using supervised data from Wall Street Journal (WSJ), Aurora-4 along with CHiME-4 real noisy data as unlabeled data. The multiple-hypothesis approach yields a relative reduction of 3.3% WER on the CHiME-4's single-channel real noisy evaluation set when compared with the single-hypothesis approach.

Transformer-based Streaming ASR with Cumulative Attention

In this paper, we propose an online attention mechanism, known as cumulative attention (CA), for streaming Transformer-based automatic speech recognition (ASR). Inspired by monotonic chunkwise attention (MoChA) and head-synchronous decoder-end adaptive computation steps (HS-DACS) algorithms, CA triggers the ASR outputs based on the acoustic information accumulated at each encoding timestep, where the decisions are made using a trainable device, referred to as halting selector. In CA, all the attention heads of the same decoder layer are synchronised to have a unified halting position. This feature effectively alleviates the problem caused by the distinct behaviour of individual heads, which may otherwise give rise to severe latency issues as encountered by MoChA. The ASR experiments conducted on AIShell-1 and Librispeech datasets demonstrate that the proposed CA-based Transformer system can achieve on par or better performance with significant reduction in latency during inference, when compared to other streaming Transformer systems in literature.

Head-synchronous Decoding for Transformer-based Streaming ASR

Online Transformer-based automatic speech recognition (ASR) systems have been extensively studied due to the increasing demand for streaming applications. Recently proposed Decoder-end Adaptive Computation Steps (DACS) algorithm for online Transformer ASR was shown to achieve state-of-the-art performance and outperform other existing methods. However, like any other online approach, the DACS-based attention heads in each of the Transformer decoder layers operate independently (or asynchronously) and lead to diverged attending positions. Since DACS employs a truncation threshold to determine the halting position, some of the attention weights are cut off untimely and might impact the stability and precision of decoding. To overcome these issues, here we propose a head-synchronous (HS) version of the DACS algorithm, where the boundary of attention is jointly detected by all the DACS heads in each decoder layer. ASR experiments on Wall Street Journal (WSJ), AIShell-1 and Librispeech show that the proposed method consistently outperforms vanilla DACS and achieves state-of-the-art performance. We will also demonstrate that HS-DACS has reduced decoding cost when compared to vanilla DACS.

End-to-end Speech Recognition with Adaptive Computation Steps

In this paper, we present Adaptive Computation Steps (ACS) algo-rithm, which enables end-to-end speech recognition models to dy-namically decide how many frames should be processed to predict a linguistic output. The model that applies ACS algorithm follows the encoder-decoder framework, while unlike the attention-based mod-els, it produces alignments independently at the encoder side using the correlation between adjacent frames. Thus, predictions can be made as soon as sufficient acoustic information is received, which makes the model applicable in online cases. Besides, a small change is made to the decoding stage of the encoder-decoder framework, which allows the prediction to exploit bidirectional contexts. We verify the ACS algorithm on a Mandarin speech corpus AIShell-1, and it achieves a 31.2% CER in the online occasion, compared to the 32.4% CER of the attention-based model. To fully demonstrate the advantage of ACS algorithm, offline experiments are conducted, in which our ACS model achieves an 18.7% CER, outperforming the attention-based counterpart with the CER of 22.0%.