Abstract Despite being the best known objective for learning speech representations, the HuBERT objective has not been further developed and improved. We argue that it is the lack of an underlying principle that stalls the development, and, in this paper, we show that predictive coding under a variational view is the principle behind the HuBERT objective. Due to its generality, our formulation provides opportunities to improve parameterization and optimization, and we show two simple modifications that bring immediate improvements to the HuBERT objective. In addition, the predictive coding formulation has tight connections to various other objectives, such as APC, CPC, wav2vec, and BEST-RQ. Empirically, the improvement in pre-training brings significant improvements to four downstream tasks: phone classification, f0 tracking, speaker recognition, and automatic speech recognition, highlighting the importance of the predictive coding interpretation.

There is an increasing interest in obtaining accurate word-level timestamps from strong automatic speech recognizers, in particular Whisper. Existing approaches either require additional training or are simply not competitive. The evaluation in prior work is also relatively loose, typically using a tolerance of more than 200 ms. In this work, we discover attention heads in Whisper that capture accurate word alignments and are distinctively different from those that do not. Moreover, we find that using characters produces finer and more accurate alignments than using wordpieces. Based on these findings, we propose an unsupervised approach to extracting word alignments by filtering attention heads while teacher forcing Whisper with characters. Our approach not only does not require training but also produces word alignments that are more accurate than prior work under a stricter tolerance between 20 ms and $100 \mathrm{~ms}$.11The source code is available at https://github.com/30stomercury/whisper-char-alignment

Existing studies on self-supervised speech representation learning have focused on developing new training methods and applying pre-trained models for different applications. However, the quality of these models is often measured by the performance of different downstream tasks. How well the representations access the information of interest is less studied. In this work, we take a closer look into existing self-supervised methods of speech from an information-theoretic perspective. We aim to develop metrics using mutual information to help practical problems such as model design and selection. We use linear probes to estimate the mutual information between the target information and learned representations, showing another insight into the accessibility to the target information from speech representations. Further, we explore the potential of evaluating representations in a self-supervised fashion, where we estimate the mutual information between different parts of the data without using any labels. Finally, we show that both supervised and unsupervised measures echo the performance of the models on layer-wise linear probing and speech recognition.

Representing speech with discrete units has been widely used in speech codec and speech generation. However, there are several unverified claims about self-supervised discrete units, such as disentangling phonetic and speaker information with k-means, or assuming information loss after k-means. In this work, we take an information-theoretic perspective to answer how much information is present (information completeness) and how much information is accessible (information accessibility), before and after residual vector quantization. We show a lower bound for information completeness and estimate completeness on discretized HuBERT representations after residual vector quantization. We find that speaker information is sufficiently present in HuBERT discrete units, and that phonetic information is sufficiently present in the residual, showing that vector quantization does not achieve disentanglement. Our results offer a comprehensive assessment on the choice of discrete units, and suggest that a lot more information in the residual should be mined rather than discarded.

SpeechBrain is an open-source Conversational AI toolkit based on PyTorch, focused particularly on speech processing tasks such as speech recognition, speech enhancement, speaker recognition, text-to-speech, and much more. It promotes transparency and replicability by releasing both the pre-trained models and the complete recipes of code and algorithms required for training them. This paper presents SpeechBrain 1.0, a significant milestone in the evolution of the toolkit, which now has over 200 recipes for speech, audio, and language processing tasks, and more than 100 models available on Hugging Face. SpeechBrain 1.0 introduces new technologies to support diverse learning modalities, Large Language Model (LLM) integration, and advanced decoding strategies, along with novel models, tasks, and modalities. It also includes a new benchmark repository, offering researchers a unified platform for evaluating models across diverse tasks.

While discrete latent variable models have had great success in self-supervised learning, most models assume that frames are independent. Due to the segmental nature of phonemes in speech perception, modeling dependencies among latent variables at the frame level can potentially improve the learned representations on phonetic-related tasks. In this work, we assume Markovian dependencies among latent variables, and propose to learn speech representations with neural hidden Markov models. Our general framework allows us to compare to self-supervised models that assume independence, while keeping the number of parameters fixed. The added dependencies improve the accessibility of phonetic information, phonetic segmentation, and the cluster purity of phones, showcasing the benefit of the assumed dependencies.

Recently, diffusion models (DMs) have been increasingly used in audio processing tasks, including speech super-resolution (SR), which aims to restore high-frequency content given low-resolution speech utterances. This is commonly achieved by conditioning the network of noise predictor with low-resolution audio. In this paper, we propose a novel sampling algorithm that communicates the information of the low-resolution audio via the reverse sampling process of DMs. The proposed method can be a drop-in replacement for the vanilla sampling process and can significantly improve the performance of the existing works. Moreover, by coupling the proposed sampling method with an unconditional DM, i.e., a DM with no auxiliary inputs to its noise predictor, we can generalize it to a wide range of SR setups. We also attain state-of-the-art results on the VCTK Multi-Speaker benchmark with this novel formulation.

With the rapid proliferation of online media sources and published news, headlines have become increasingly important for attracting readers to news articles, since users may be overwhelmed with the massive information. In this paper, we generate inspired headlines that preserve the nature of news articles and catch the eye of the reader simultaneously. The task of inspired headline generation can be viewed as a specific form of Headline Generation (HG) task, with the emphasis on creating an attractive headline from a given news article. To generate inspired headlines, we propose a novel framework called POpularity-Reinforced Learning for inspired Headline Generation (PORL-HG). PORL-HG exploits the extractive-abstractive architecture with 1) Popular Topic Attention (PTA) for guiding the extractor to select the attractive sentence from the article and 2) a popularity predictor for guiding the abstractor to rewrite the attractive sentence. Moreover, since the sentence selection of the extractor is not differentiable, techniques of reinforcement learning (RL) are utilized to bridge the gap with rewards obtained from a popularity score predictor. Through quantitative and qualitative experiments, we show that the proposed PORL-HG significantly outperforms the state-of-the-art headline generation models in terms of attractiveness evaluated by both human (71.03%) and the predictor (at least 27.60%), while the faithfulness of PORL-HG is also comparable to the state-of-the-art generation model.

Developing a robust emotion speech recognition (SER) system for human dialog is important in advancing conversational agent design. In this paper, we proposed a novel inference algorithm, a dialogical emotion decoding (DED) algorithm, that treats a dialog as a sequence and consecutively decode the emotion states of each utterance over time with a given recognition engine. This decoder is trained by incorporating intra- and inter-speakers emotion influences within a conversation. Our approach achieves a 70.1% in four class emotion on the IEMOCAP database, which is 3% over the state-of-art model. The evaluation is further conducted on a multi-party interaction database, the MELD, which shows a similar effect. Our proposed DED is in essence a conversational emotion rescoring decoder that can also be flexibly combined with different SER engines.

Obtaining robust speech emotion recognition (SER) in scenarios of spoken interactions is critical to the developments of next generation human-machine interface. Previous research has largely focused on performing SER by modeling each utterance of the dialog in isolation without considering the transactional and dependent nature of the human-human conversation. In this work, we propose an interaction-aware attention network (IAAN) that incorporate contextual information in the learned vocal representation through a novel attention mechanism. Our proposed method achieves 66.3% accuracy (7.9% over baseline methods) in four class emotion recognition and is also the current state-of-art recognition rates obtained on the benchmark database.