Compression-based representations (CBRs) from neural audio codecs such as EnCodec capture intricate acoustic features like pitch and timbre, while representation-learning-based representations (RLRs) from pre-trained models trained for speech representation learning such as WavLM encode high-level semantic and prosodic information. Previous research on Speech Emotion Recognition (SER) has explored both, however, fusion of CBRs and RLRs haven't been explored yet. In this study, we solve this gap and investigate the fusion of RLRs and CBRs and hypothesize they will be more effective by providing complementary information. To this end, we propose, HYFuse, a novel framework that fuses the representations by transforming them to hyperbolic space. With HYFuse, through fusion of x-vector (RLR) and Soundstream (CBR), we achieve the top performance in comparison to individual representations as well as the homogeneous fusion of RLRs and CBRs and report SOTA.
Recent advancements in speech encoders have drawn attention due to their integration with Large Language Models for various speech tasks. While most research has focused on either causal or full-context speech encoders, there's limited exploration to effectively handle both streaming and non-streaming applications, while achieving state-of-the-art performance. We introduce DuRep, a Dual-mode Speech Representation learning setup, which enables a single speech encoder to function efficiently in both offline and online modes without additional parameters or mode-specific adjustments, across downstream tasks. DuRep-200M, our 200M parameter dual-mode encoder, achieves 12% and 11.6% improvements in streaming and non-streaming modes, over baseline encoders on Multilingual ASR. Scaling this approach to 2B parameters, DuRep-2B sets new performance benchmarks across ASR and non-ASR tasks. Our analysis reveals interesting trade-offs between acoustic and semantic information across encoder layers.
The use of audio recordings of human speech to train LLMs poses privacy concerns due to these models' potential to generate outputs that closely resemble artifacts in the training data. In this study, we propose a speaker privacy-preserving representation learning method through the Universal Speech Codec (USC), a computationally efficient encoder-decoder model that disentangles speech into: (i) privacy-preserving semantically rich representations, capturing content and speech paralinguistics, and (ii) residual acoustic and speaker representations that enables high-fidelity reconstruction. Extensive evaluations presented show that USC's semantic representation preserves content, prosody, and sentiment, while removing potentially identifiable speaker attributes. Combining both representations, USC achieves state-of-the-art speech reconstruction. Additionally, we introduce an evaluation methodology for measuring privacy-preserving properties, aligning with perceptual tests. We compare USC against other codecs in the literature and demonstrate its effectiveness on privacy-preserving representation learning, illustrating the trade-offs of speaker anonymization, paralinguistics retention and content preservation in the learned semantic representations. Audio samples are shared in https://www.amazon.science/usc-samples.
Audio-visual speech recognition (AVSR) incorporates auditory and visual modalities to improve recognition accuracy, particularly in noisy environments where audio-only speech systems are insufficient. While previous research has largely addressed audio disruptions, few studies have dealt with visual corruptions, e.g., lip occlusions or blurred videos, which are also detrimental. To address this real-world challenge, we propose CAV2vec, a novel self-supervised speech representation learning framework particularly designed to handle audio-visual joint corruption. CAV2vec employs a self-distillation approach with a corrupted prediction task, where the student model learns to predict clean targets, generated by the teacher model, with corrupted input frames. Specifically, we suggest a unimodal multi-task learning, which distills cross-modal knowledge and aligns the corrupted modalities, by predicting clean audio targets with corrupted videos, and clean video targets with corrupted audios. This strategy mitigates the dispersion in the representation space caused by corrupted modalities, leading to more reliable and robust audio-visual fusion. Our experiments on robust AVSR benchmarks demonstrate that the corrupted representation learning method significantly enhances recognition accuracy across generalized environments involving various types of corruption. Our code is available at https://github.com/sungnyun/cav2vec.
Pre-training and representation learning have been playing an increasingly important role in modern speech processing. Nevertheless, different applications have been relying on different foundation models, since predominant pre-training techniques are either designed for discriminative tasks or generative tasks. In this work, we make the first attempt at building a unified pre-training framework for both types of tasks in speech. We show that with the appropriate design choices for pre-training, one can jointly learn a representation encoder and generative audio decoder that can be applied to both types of tasks. We propose UniWav, an encoder-decoder framework designed to unify pre-training representation learning and generative tasks. On speech recognition, text-to-speech, and speech tokenization, UniWav achieves comparable performance to different existing foundation models, each trained on a specific task. Our findings suggest that a single general-purpose foundation model for speech can be built to replace different foundation models, reducing the overhead and cost of pre-training.
Self-supervised learning (SSL) has achieved great success in speech-related tasks. While Transformer and Conformer architectures have dominated SSL backbones, encoders like Zipformer, which excel in automatic speech recognition (ASR), remain unexplored in SSL. Concurrently, inefficiencies in data processing within existing SSL training frameworks, such as fairseq, pose challenges in managing the growing volumes of training data. To address these issues, we propose k2SSL, an open-source framework that offers faster, more memory-efficient, and better-performing self-supervised speech representation learning, focusing on downstream ASR tasks. The optimized HuBERT and proposed Zipformer-based SSL systems exhibit substantial reductions in both training time and memory usage during SSL training. Experiments on LibriSpeech demonstrate that Zipformer Base significantly outperforms HuBERT and WavLM, achieving up to a 34.8% relative WER reduction compared to HuBERT Base after fine-tuning, along with a 3.5x pre-training speedup in GPU hours. When scaled to 60k hours of LibriLight data, Zipformer Large exhibits remarkable efficiency, matching HuBERT Large's performance while requiring only 5/8 pre-training steps.
In this paper, we present EH-MAM (Easy-to-Hard adaptive Masked Acoustic Modeling), a novel self-supervised learning approach for speech representation learning. In contrast to the prior methods that use random masking schemes for Masked Acoustic Modeling (MAM), we introduce a novel selective and adaptive masking strategy. Specifically, during SSL training, we progressively introduce harder regions to the model for reconstruction. Our approach automatically selects hard regions and is built on the observation that the reconstruction loss of individual frames in MAM can provide natural signals to judge the difficulty of solving the MAM pre-text task for that frame. To identify these hard regions, we employ a teacher model that first predicts the frame-wise losses and then decides which frames to mask. By learning to create challenging problems, such as identifying harder frames and solving them simultaneously, the model is able to learn more effective representations and thereby acquire a more comprehensive understanding of the speech. Quantitatively, EH-MAM outperforms several state-of-the-art baselines across various low-resource speech recognition and SUPERB benchmarks by 5%-10%. Additionally, we conduct a thorough analysis to show that the regions masked by EH-MAM effectively capture useful context across speech frames.
Information in speech can be divided into two categories: what is being said (content) and how it is expressed (other). Current state-of-the-art (SOTA) techniques model speech at fixed segments, usually 10-25 ms, using a single embedding. Given the orthogonal nature of other and content information, attempting to optimize both within a single embedding results in suboptimal solutions. This approach divides the models capacity, limiting its ability to build complex hierarchical features effectively. In this work, we present an end-to-end speech representation learning framework designed to jointly optimize the other and content information (JOOCI) in speech. By using separate learnable parameters, JOOCI addresses this optimization challenge by modeling other and content information independently. Our results show that JOOCI consistently outperforms other SOTA models of similar size (100 million parameters) and pre-training data used (960 hours) by a significant margin when evaluated on a range of speech downstream tasks in the SUPERB benchmark, as shown in Table 1.
In this study, for the first time, we extensively investigate whether music foundation models (MFMs) or speech foundation models (SFMs) work better for singing voice deepfake detection (SVDD), which has recently attracted attention in the research community. For this, we perform a comprehensive comparative study of state-of-the-art (SOTA) MFMs (MERT variants and music2vec) and SFMs (pre-trained for general speech representation learning as well as speaker recognition). We show that speaker recognition SFM representations perform the best amongst all the foundation models (FMs), and this performance can be attributed to its higher efficacy in capturing the pitch, tone, intensity, etc, characteristics present in singing voices. To our end, we also explore the fusion of FMs for exploiting their complementary behavior for improved SVDD, and we propose a novel framework, FIONA for the same. With FIONA, through the synchronization of x-vector (speaker recognition SFM) and MERT-v1-330M (MFM), we report the best performance with the lowest Equal Error Rate (EER) of 13.74 %, beating all the individual FMs as well as baseline FM fusions and achieving SOTA results.
Self-supervised speech representation learning has become essential for extracting meaningful features from untranscribed audio. Recent advances highlight the potential of deriving discrete symbols from the features correlated with linguistic units, which enables text-less training across diverse tasks. In particular, sentence-level Self-Distillation of the pretrained HuBERT (SD-HuBERT) induces syllabic structures within latent speech frame representations extracted from an intermediate Transformer layer. In SD-HuBERT, sentence-level representation is accumulated from speech frame features through self-attention layers using a special CLS token. However, we observe that the information aggregated in the CLS token correlates more with speaker identity than with linguistic content. To address this, we propose a speech-only self-supervised fine-tuning approach that separates syllabic units from speaker information. Our method introduces speaker perturbation as data augmentation and adopts a frame-level training objective to prevent the CLS token from aggregating paralinguistic information. Experimental results show that our approach surpasses the current state-of-the-art method in most syllable segmentation and syllabic unit quality metrics on Librispeech, underscoring its effectiveness in promoting syllabic organization within speech-only models.
Self-supervised learning (SSL) has garnered significant attention in speech processing, excelling in linguistic tasks such as speech recognition. However, jointly improving the performance of pre-trained models on various downstream tasks, each requiring different speech information, poses significant challenges. To this purpose, we propose a progressive residual extraction based self-supervised learning method, named ProgRE. Specifically, we introduce two lightweight and specialized task modules into an encoder-style SSL backbone to enhance its ability to extract pitch variation and speaker information from speech. Furthermore, to prevent the interference of reinforced pitch variation and speaker information with irrelevant content information learning, we residually remove the information extracted by these two modules from the main branch. The main branch is then trained using HuBERT's speech masking prediction to ensure the performance of the Transformer's deep-layer features on content tasks. In this way, we can progressively extract pitch variation, speaker, and content representations from the input speech. Finally, we can combine multiple representations with diverse speech information using different layer weights to obtain task-specific representations for various downstream tasks. Experimental results indicate that our proposed method achieves joint performance improvements on various tasks, such as speaker identification, speech recognition, emotion recognition, speech enhancement, and voice conversion, compared to excellent SSL methods such as wav2vec2.0, HuBERT, and WavLM.
Speech modeling methods learn one embedding for a fixed segment of speech, typically in between 10-25 ms. The information present in speech can be divided into two categories: "what is being said" (content) and "how it is expressed" (other) and these two are orthogonal in nature causing the optimization algorithm to find a sub-optimal solution if forced to optimize together. This leads to sub-optimal performance in one or all downstream tasks as shown by previous studies. Current self-supervised learning (SSL) methods such as HuBERT are very good at modeling the content information present in speech. Data augmentation improves the performance on tasks which require effective modeling of other information but this leads to a divided capacity of the model. In this work, we conduct a preliminary study to understand the importance of modeling other information using separate learnable parameters. We propose a modified version of HuBERT, termed Other HuBERT (O-HuBERT), to test our hypothesis. Our findings are twofold: first, the O-HuBERT method is able to utilize all layers to build complex features to encode other information; second, a robust data augmentation strategy is essential for learning the information required by tasks that depend on other information and to achieve state-of-the-art (SOTA) performance on the SUPERB benchmark with a similarly sized model (100 million parameters) and pre-training data (960 hours).
The popular frameworks for self-supervised learning of speech representations have largely focused on frame-level masked prediction of speech regions. While this has shown promising downstream task performance for speech recognition and related tasks, this has largely ignored factors of speech that are encoded at coarser level, like characteristics of the speaker or channel that remain consistent through-out a speech utterance. In this work, we propose a framework for Learning Disentangled Self Supervised (termed as Learn2Diss) representations of speech, which consists of frame-level and an utterance-level encoder modules. The two encoders are initially learned independently, where the frame-level model is largely inspired by existing self supervision techniques, thereby learning pseudo-phonemic representations, while the utterance-level encoder is inspired by constrastive learning of pooled embeddings, thereby learning pseudo-speaker representations. The joint learning of these two modules consists of disentangling the two encoders using a mutual information based criterion. With several downstream evaluation experiments, we show that the proposed Learn2Diss achieves state-of-the-art results on a variety of tasks, with the frame-level encoder representations improving semantic tasks, while the utterance-level representations improve non-semantic tasks.
Self-supervised learning (SSL) has helped extend speech technologies to more languages by reducing the need for labeled data. However, models are still far from supporting the world's 7000+ languages. We propose XEUS, a Cross-lingual Encoder for Universal Speech, trained on over 1 million hours of data across 4057 languages, extending the language coverage of SSL models 4-fold. We combine 1 million hours of speech from existing publicly accessible corpora with a newly created corpus of 7400+ hours from 4057 languages, which will be publicly released. To handle the diverse conditions of multilingual speech data, we augment the typical SSL masked prediction approach with a novel dereverberation objective, increasing robustness. We evaluate XEUS on several benchmarks, and show that it consistently outperforms or achieves comparable results to state-of-the-art (SOTA) SSL models across a variety of tasks. XEUS sets a new SOTA on the ML-SUPERB benchmark: it outperforms MMS 1B and w2v-BERT 2.0 v2 by 0.8% and 4.4% respectively, despite having less parameters or pre-training data. Checkpoints, code, and data are found in https://www.wavlab.org/activities/2024/xeus/.
Speech Self-Supervised Learning (SSL) has demonstrated considerable efficacy in various downstream tasks. Nevertheless, prevailing self-supervised models often overlook the incorporation of emotion-related prior information, thereby neglecting the potential enhancement of emotion task comprehension through emotion prior knowledge in speech. In this paper, we propose an emotion-aware speech representation learning with intensity knowledge. Specifically, we extract frame-level emotion intensities using an established speech-emotion understanding model. Subsequently, we propose a novel emotional masking strategy (EMS) to incorporate emotion intensities into the masking process. We selected two representative models based on Transformer and CNN, namely MockingJay and Non-autoregressive Predictive Coding (NPC), and conducted experiments on IEMOCAP dataset. Experiments have demonstrated that the representations derived from our proposed method outperform the original model in SER task.
We present mHuBERT-147, the first general-purpose massively multilingual HuBERT speech representation model trained on 90K hours of clean, open-license data. To scale up the multi-iteration HuBERT approach, we use faiss-based clustering, achieving 5.2x faster label assignment than the original method. We also apply a new multilingual batching up-sampling strategy, leveraging both language and dataset diversity. After 3 training iterations, our compact 95M parameter mHuBERT-147 outperforms larger models trained on substantially more data. We rank second and first on the ML-SUPERB 10min and 1h leaderboards, with SOTA scores for 3 tasks. Across ASR/LID tasks, our model consistently surpasses XLS-R (300M params; 436K hours) and demonstrates strong competitiveness against the much larger MMS (1B params; 491K hours). Our findings indicate that mHuBERT-147 is a promising model for multilingual speech tasks, offering an unprecedented balance between high performance and parameter efficiency.
Speech recognition and translation systems perform poorly on noisy inputs, which are frequent in realistic environments. Augmenting these systems with visual signals has the potential to improve robustness to noise. However, audio-visual (AV) data is only available in limited amounts and for fewer languages than audio-only resources. To address this gap, we present XLAVS-R, a cross-lingual audio-visual speech representation model for noise-robust speech recognition and translation in over 100 languages. It is designed to maximize the benefits of limited multilingual AV pre-training data, by building on top of audio-only multilingual pre-training and simplifying existing pre-training schemes. Extensive evaluation on the MuAViC benchmark shows the strength of XLAVS-R on downstream audio-visual speech recognition and translation tasks, where it outperforms the previous state of the art by up to 18.5% WER and 4.7 BLEU given noisy AV inputs, and enables strong zero-shot audio-visual ability with audio-only fine-tuning.
Self-supervised speech representation learning enables the extraction of meaningful features from raw waveforms. These features can then be efficiently used across multiple downstream tasks. However, two significant issues arise when considering the deployment of such methods ``in-the-wild": (i) Their large size, which can be prohibitive for edge applications; and (ii) their robustness to detrimental factors, such as noise and/or reverberation, that can heavily degrade the performance of such systems. In this work, we propose RobustDistiller, a novel knowledge distillation mechanism that tackles both problems jointly. Simultaneously to the distillation recipe, we apply a multi-task learning objective to encourage the network to learn noise-invariant representations by denoising the input. The proposed mechanism is evaluated on twelve different downstream tasks. It outperforms several benchmarks regardless of noise type, or noise and reverberation levels. Experimental results show that the new Student model with 23M parameters can achieve results comparable to the Teacher model with 95M parameters. Lastly, we show that the proposed recipe can be applied to other distillation methodologies, such as the recent DPWavLM. For reproducibility, code and model checkpoints will be made available at \mbox{\url{https://github.com/Hguimaraes/robustdistiller}}.
Foundation models in speech are often trained using many GPUs, which implicitly leads to large effective batch sizes. In this paper we study the effect of batch size on pre-training, both in terms of statistics that can be monitored during training, and in the effect on the performance of a downstream fine-tuning task. By using batch sizes varying from 87.5 seconds to 80 minutes of speech we show that, for a fixed amount of iterations, larger batch sizes result in better pre-trained models. However, there is lower limit for stability, and an upper limit for effectiveness. We then show that the quality of the pre-trained model depends mainly on the amount of speech data seen during training, i.e., on the product of batch size and number of iterations. All results are produced with an independent implementation of the wav2vec 2.0 architecture, which to a large extent reproduces the results of the original work (arXiv:2006.11477). Our extensions can help researchers choose effective operating conditions when studying self-supervised learning in speech, and hints towards benchmarking self-supervision with a fixed amount of seen data. Code and model checkpoints are available at https://github.com/nikvaessen/w2v2-batch-size.
Dysarthric speech reconstruction (DSR) systems aim to automatically convert dysarthric speech into normal-sounding speech. The technology eases communication with speakers affected by the neuromotor disorder and enhances their social inclusion. NED-based (Neural Encoder-Decoder) systems have significantly improved the intelligibility of the reconstructed speech as compared with GAN-based (Generative Adversarial Network) approaches, but the approach is still limited by training inefficiency caused by the cascaded pipeline and auxiliary tasks of the content encoder, which may in turn affect the quality of reconstruction. Inspired by self-supervised speech representation learning and discrete speech units, we propose a Unit-DSR system, which harnesses the powerful domain-adaptation capacity of HuBERT for training efficiency improvement and utilizes speech units to constrain the dysarthric content restoration in a discrete linguistic space. Compared with NED approaches, the Unit-DSR system only consists of a speech unit normalizer and a Unit HiFi-GAN vocoder, which is considerably simpler without cascaded sub-modules or auxiliary tasks. Results on the UASpeech corpus indicate that Unit-DSR outperforms competitive baselines in terms of content restoration, reaching a 28.2% relative average word error rate reduction when compared to original dysarthric speech, and shows robustness against speed perturbation and noise.