AI research atlas / v2
Learn AI papers in the right order.
Start with landmark ideas, move through foundations, then branch into LLMs, GenAI, agents, systems, and safety with a reading path that keeps the field from feeling random.
Build the mental timeline before going deep.
Move from foundations to modern systems.
Learning path
Where to start, and what to read next
Orientation / 1-2 weeks
Start Here
Read the papers everyone keeps referencing so the rest of the map has anchors.
Foundations / 2-4 weeks
Classical ML
Learn the statistical and probabilistic ideas that still sit under modern models.
Foundations / 1-2 weeks
Optimization
Understand the training mechanics behind gradient-based learning.
Builder / 3-5 weeks
Deep Learning Core
Move through representation learning, CNNs, residual networks, and scaling patterns.
Builder / 3-6 weeks
Sequence Models and LLMs
Study attention, transformers, language modeling, instruction tuning, and evaluation.
Specialist / 3-6 weeks
Generative AI
Compare GANs, diffusion, autoregressive generation, and modern GenAI workflows.
Specialist / 2-4 weeks
Multimodal and Retrieval
Connect language with images, retrieval, embeddings, and real-world knowledge access.
Specialist / 3-5 weeks
RL and Agents
Learn decision making, feedback, policy learning, and agent-style systems.
Practitioner / 2-4 weeks
Systems and Scaling
Understand the infrastructure and engineering papers behind large-scale training.
Practitioner / 2-4 weeks
Safety and Interpretability
Study robustness, alignment, transparency, and how to reason about model behavior.
Learning Paradigms
Trust and Deployment
Research library
Speech and Audio
Showing papers for this learning path. Open any paper card to read the full paper and related resources.
Cross-Attention is all you need: Real-Time Streaming Transformers for Personalised Speech Enhancement
Personalised speech enhancement (PSE), which extracts only the speech of a target user and removes everything else from a recorded audio clip, can potentially improve users' experiences of audio AI modules deployed in the wild. To support a large variety of downstream audio tasks, such as real-time ASR and audio-call enhancement, a PSE solution should operate in a streaming mode, i.e., input audio cleaning should happen in real-time with a small latency and real-time factor. Personalisation is typically achieved by extracting a target speaker's voice profile from an enrolment audio, in the form of a static embedding vector, and then using it to condition the output of a PSE model. However, a fixed target speaker embedding may not be optimal under all conditions. In this work, we present a streaming Transformer-based PSE model and propose a novel cross-attention approach that gives adaptive target speaker representations. We present extensive experiments and show that our proposed cross-attention approach outperforms competitive baselines consistently, even when our model is only approximately half the size.
Dual-path Attention is All You Need for Audio-Visual Speech Extraction
No abstract available yet.
Advancing automatic speech recognition using feature fusion with self-supervised learning features: A case study on Fearless Steps Apollo corpus
Using self-supervised learning (SSL) models has significantly improved performance for downstream speech tasks, surpassing the capabilities of traditional hand-crafted features. This study investigates the amalgamation of SSL models, with the aim to leverage both their individual strengths and refine extracted features to achieve improved speech recognition models for naturalistic scenarios. Our research investigates the massive naturalistic Fearless Steps (FS) APOLLO resource, with particular focus on the FS Challenge (FSC) Phase-4 corpus, providing the inaugural analysis of this dataset. Additionally, we incorporate the CHiME-6 dataset to evaluate performance across diverse naturalistic speech scenarios. While exploring previously proposed Feature Refinement Loss and fusion methods, we found these methods to be less effective on the FSC Phase-4 corpus. To address this, we introduce a novel deep cross-attention (DCA) fusion method, designed to elevate performance, especially for the FSC Phase-4 corpus. Our objective is to foster creation of superior FS APOLLO community resources, catering to the diverse needs of researchers across various disciplines. The proposed solution achieves an absolute +1.1% improvement in WER, providing effective meta-data creation for the massive FS APOLLO community resource.
From Alignment to Advancement: Bootstrapping Audio-Language Alignment with Synthetic Data
Audio-aware large language models (ALLMs) have recently made great strides in understanding and processing audio inputs. These models are typically adapted from text-based large language models (LLMs) through additional training on audio-related tasks. This adaptation process presents two major limitations. First, ALLMs often suffer from catastrophic forgetting, where crucial textual capabilities like instruction-following are lost after training on audio data. In some cases, models may even hallucinate sounds that are not present in the input audio, raising concerns about reliability. Second, achieving cross-modal alignment between audio and language typically relies on large collections of task-specific question-answer pairs for instruction tuning, making it resource-intensive. To address these issues, previous works have leveraged the backbone LLMs to synthesize general-purpose, caption-style alignment data. In this paper, we propose a data generation framework that produces contrastive-like training data, designed to enhance ALLMs' ability to differentiate between present and absent sounds. We further extend our approach to multi-audio scenarios, enabling the model to either explain differences between audio inputs or produce unified captions that describe all inputs, thereby enhancing audio-language alignment. We refer to the entire ALLM training framework as bootstrapping audio-language alignment via synthetic data generation from backbone LLMs (BALSa). Experimental results indicate that our method effectively mitigates audio hallucinations while reliably maintaining strong performance on audio understanding and reasoning benchmarks, as well as instruction-following skills. Moreover, incorporating multi-audio training further enhances the model's comprehension and reasoning capabilities. Overall, BALSa offers an efficient and scalable approach to developing ALLMs.
Unified Cross-modal Translation of Score Images, Symbolic Music, and Performance Audio
Music exists in various modalities, such as score images, symbolic scores, MIDI, and audio. Translations between each modality are established as core tasks of music information retrieval, such as automatic music transcription (audio-to-MIDI) and optical music recognition (score image to symbolic score). However, most past work on multimodal translation trains specialized models on individual translation tasks. In this paper, we propose a unified approach, where we train a general-purpose model on many translation tasks simultaneously. Two key factors make this unified approach viable: a new large-scale dataset and the tokenization of each modality. Firstly, we propose a new dataset that consists of more than 1,300 hours of paired audio-score image data collected from YouTube videos, which is an order of magnitude larger than any existing music modal translation datasets. Secondly, our unified tokenization framework discretizes score images, audio, MIDI, and MusicXML into a sequence of tokens, enabling a single encoder-decoder Transformer to tackle multiple cross-modal translation as one coherent sequence-to-sequence task. Experimental results confirm that our unified multitask model improves upon single-task baselines in several key areas, notably reducing the symbol error rate for optical music recognition from 24.58% to a state-of-the-art 13.67%, while similarly substantial improvements are observed across the other translation tasks. Notably, our approach achieves the first successful score-image-conditioned audio generation, marking a significant breakthrough in cross-modal music generation.
dCoNNear: An Artifact-Free Neural Network Architecture for Closed-loop Audio Signal Processing
Recent advances in deep neural networks (DNNs) have significantly improved various audio processing applications, including speech enhancement, synthesis, and hearing-aid algorithms. DNN-based closed-loop systems have gained popularity in these applications due to their robust performance and ability to adapt to diverse conditions. Despite their effectiveness, current DNN-based closed-loop systems often suffer from sound quality degradation caused by artifacts introduced by suboptimal sampling methods. To address this challenge, we introduce dCoNNear, a novel DNN architecture designed for seamless integration into closed-loop frameworks. This architecture specifically aims to prevent the generation of spurious artifacts-most notably tonal and aliasing artifacts arising from non-ideal sampling layers. We demonstrate the effectiveness of dCoNNear through a proof-of-principle example within a closed-loop framework that employs biophysically realistic models of auditory processing for both normal and hearing-impaired profiles to design personalized hearing-aid algorithms. We further validate the broader applicability and artifact-free performance of dCoNNear through speech-enhancement experiments, confirming its ability to improve perceptual sound quality without introducing architecture-induced artifacts. Our results show that dCoNNear not only accurately simulates all processing stages of existing non-DNN biophysical models but also significantly improves sound quality by eliminating audible artifacts in both hearing-aid and speech-enhancement applications. This study offers a robust, perceptually transparent closed-loop processing framework for high-fidelity audio applications.
Adaptive Convolution for CNN-based Speech Enhancement Models
Deep learning-based speech enhancement methods have significantly improved speech quality and intelligibility. Convolutional neural networks (CNNs) have been proven to be essential components of many high-performance models. In this paper, we introduce adaptive convolution, an efficient and versatile convolutional module that enhances the model's capability to adaptively represent speech signals. Adaptive convolution performs frame-wise causal dynamic convolution, generating time-varying kernels for each frame by assembling multiple parallel candidate kernels. A lightweight attention mechanism is proposed for adaptive convolution, leveraging both current and historical information to assign adaptive weights to each candidate kernel. This enables the convolution operation to adapt to frame-level speech spectral features, leading to more efficient extraction and reconstruction. We integrate adaptive convolution into various CNN-based models, highlighting its generalizability. Experimental results demonstrate that adaptive convolution significantly improves the performance with negligible increases in computational complexity, especially for lightweight models. Moreover, we present an intuitive analysis revealing a strong correlation between kernel selection and signal characteristics. Furthermore, we propose the adaptive convolutional recurrent network (AdaptCRN), an ultra-lightweight model that incorporates adaptive convolution and an efficient encoder-decoder design, achieving superior performance compared to models with similar or even higher computational costs.
Interpreting the Role of Visemes in Audio-Visual Speech Recognition
Audio-Visual Speech Recognition (AVSR) models have surpassed their audio-only counterparts in terms of performance. However, the interpretability of AVSR systems, particularly the role of the visual modality, remains under-explored. In this paper, we apply several interpretability techniques to examine how visemes are encoded in AV-HuBERT a state-of-the-art AVSR model. First, we use t-distributed Stochastic Neighbour Embedding (t-SNE) to visualize learned features, revealing natural clustering driven by visual cues, which is further refined by the presence of audio. Then, we employ probing to show how audio contributes to refining feature representations, particularly for visemes that are visually ambiguous or under-represented. Our findings shed light on the interplay between modalities in AVSR and could point to new strategies for leveraging visual information to improve AVSR performance.
MambAttention: Mamba with Multi-Head Attention for Generalizable Single-Channel Speech Enhancement
With new sequence models like Mamba and xLSTM, several studies have shown that these models match or outperform the state-of-the-art in single-channel speech enhancement and audio representation learning. However, prior research has demonstrated that sequence models like LSTM and Mamba tend to overfit to the training set. To address this, previous works have shown that adding self-attention to LSTMs substantially improves generalization performance for single-channel speech enhancement. Nevertheless, neither the concept of hybrid Mamba and time-frequency attention models nor their generalization performance have been explored for speech enhancement. In this paper, we propose a novel hybrid architecture, MambAttention, which combines Mamba and shared time- and frequency-multi-head attention modules for generalizable single-channel speech enhancement. To train our model, we introduce VB-DemandEx, a dataset inspired by VoiceBank+Demand but with more challenging noise types and lower signal-to-noise ratios. Trained on VB-DemandEx, MambAttention significantly outperforms existing state-of-the-art discriminative LSTM-, xLSTM-, Mamba-, and Conformer-based systems of similar complexity across all reported metrics on two out-of-domain datasets: DNS 2020 without reverberation and EARS-WHAM_v2. MambAttention also matches or outperforms generative diffusion models in generalization performance while being competitive with language model baselines. Ablation studies highlight the importance of weight sharing between time- and frequency-multi-head attention modules for generalization performance. Finally, we explore integrating the shared time- and frequency-multi-head attention modules with LSTM and xLSTM, which yields a notable performance improvement on the out-of-domain datasets. Yet, MambAttention remains superior for cross-corpus generalization across all reported evaluation metrics.
SpeechPrompt: Prompting Speech Language Models for Speech Processing Tasks
Prompting has become a practical method for utilizing pre-trained language models (LMs). This approach offers several advantages. It allows an LM to adapt to new tasks with minimal training and parameter updates, thus achieving efficiency in both storage and computation. Additionally, prompting modifies only the LM's inputs and harnesses the generative capabilities of language models to address various downstream tasks in a unified manner. This significantly reduces the need for human labor in designing task-specific models. These advantages become even more evident as the number of tasks served by the LM scales up. Motivated by the strengths of prompting, we are the first to explore the potential of prompting speech LMs in the domain of speech processing. Recently, there has been a growing interest in converting speech into discrete units for language modeling. Our pioneer research demonstrates that these quantized speech units are highly versatile within our unified prompting framework. Not only can they serve as class labels, but they also contain rich phonetic information that can be re-synthesized back into speech signals for speech generation tasks. Specifically, we reformulate speech processing tasks into speech-to-unit generation tasks. As a result, we can seamlessly integrate tasks such as speech classification, sequence generation, and speech generation within a single, unified prompting framework. The experiment results show that the prompting method can achieve competitive performance compared to the strong fine-tuning method based on self-supervised learning models with a similar number of trainable parameters. The prompting method also shows promising results in the few-shot setting. Moreover, with the advanced speech LMs coming into the stage, the proposed prompting framework attains great potential.
Towards Improved Objective Perceptual Audio Quality Assessment -- Part 1: A Novel Data-Driven Cognitive Model
Efficient audio quality assessment is vital for streamlining audio codec development. Objective assessment tools have been developed over time to algorithmically predict quality ratings from subjective assessments, the gold standard for quality judgment. Many of these tools use perceptual auditory models to extract audio features that are mapped to a basic audio quality score prediction using machine learning algorithms and subjective scores as training data. However, existing tools struggle with generalization in quality prediction, especially when faced with unknown signal and distortion types. This is particularly evident in the presence of signals coded using non-waveform-preserving parametric techniques. Addressing these challenges, this two-part work proposes extensions to the Perceptual Evaluation of Audio Quality (PEAQ - ITU-R BS.1387-1) recommendation. Part 1 focuses on increasing generalization, while Part 2 targets accurate spatial audio quality measurement in audio coding. To enhance prediction generalization, this paper (Part 1) introduces a novel machine learning approach that uses subjective data to model cognitive aspects of audio quality perception. The proposed method models the perceived severity of audible distortions by adaptively weighting different distortion metrics. The weights are determined using an interaction cost function that captures relationships between distortion salience and cognitive effects. Compared to other machine learning methods and established tools, the proposed architecture achieves higher prediction accuracy on large databases of previously unseen subjective quality scores. The perceptually-motivated model offers a more manageable alternative to general-purpose machine learning algorithms, allowing potential extensions and improvements to multi-dimensional quality measurement without complete retraining.
Speech Separation with Pretrained Frontend to Minimize Domain Mismatch
Speech separation seeks to separate individual speech signals from a speech mixture. Typically, most separation models are trained on synthetic data due to the unavailability of target reference in real-world cocktail party scenarios. As a result, there exists a domain gap between real and synthetic data when deploying speech separation models in real-world applications. In this paper, we propose a self-supervised domain-invariant pretrained (DIP) frontend that is exposed to mixture data without the need for target reference speech. The DIP frontend utilizes a Siamese network with two innovative pretext tasks, mixture predictive coding (MPC) and mixture invariant coding (MIC), to capture shared contextual cues between real and synthetic unlabeled mixtures. Subsequently, we freeze the DIP frontend as a feature extractor when training the downstream speech separation models on synthetic data. By pretraining the DIP frontend with the contextual cues, we expect that the speech separation skills learned from synthetic data can be effectively transferred to real data. To benefit from the DIP frontend, we introduce a novel separation pipeline to align the feature resolution of the separation models. We evaluate the speech separation quality on standard benchmarks and real-world datasets. The results confirm the superiority of our DIP frontend over existing speech separation models. This study underscores the potential of large-scale pretraining to enhance the quality and intelligibility of speech separation in real-world applications.
CochCeps-Augment: A Novel Self-Supervised Contrastive Learning Using Cochlear Cepstrum-based Masking for Speech Emotion Recognition
Self-supervised learning (SSL) for automated speech recognition in terms of its emotional content, can be heavily degraded by the presence noise, affecting the efficiency of modeling the intricate temporal and spectral informative structures of speech. Recently, SSL on large speech datasets, as well as new audio-specific SSL proxy tasks, such as, temporal and frequency masking, have emerged, yielding superior performance compared to classic approaches drawn from the image augmentation domain. Our proposed contribution builds upon this successful paradigm by introducing CochCeps-Augment, a novel bio-inspired masking augmentation task for self-supervised contrastive learning of speech representations. Specifically, we utilize the newly introduced bio-inspired cochlear cepstrogram (CCGRAM) to derive noise robust representations of input speech, that are then further refined through a self-supervised learning scheme. The latter employs SimCLR to generate contrastive views of a CCGRAM through masking of its angle and quefrency dimensions. Our experimental approach and validations on the emotion recognition K-EmoCon benchmark dataset, for the first time via a speaker-independent approach, features unsupervised pre-training, linear probing and fine-tuning. Our results potentiate CochCeps-Augment to serve as a standard tool in speech emotion recognition analysis, showing the added value of incorporating bio-inspired masking as an informative augmentation task for self-supervision. Our code for implementing CochCeps-Augment will be made available at: https://github.com/GiannisZgs/CochCepsAugment.
Changing Data Sources in the Age of Machine Learning for Official Statistics
Data science has become increasingly essential for the production of official statistics, as it enables the automated collection, processing, and analysis of large amounts of data. With such data science practices in place, it enables more timely, more insightful and more flexible reporting. However, the quality and integrity of data-science-driven statistics rely on the accuracy and reliability of the data sources and the machine learning techniques that support them. In particular, changes in data sources are inevitable to occur and pose significant risks that are crucial to address in the context of machine learning for official statistics. This paper gives an overview of the main risks, liabilities, and uncertainties associated with changing data sources in the context of machine learning for official statistics. We provide a checklist of the most prevalent origins and causes of changing data sources; not only on a technical level but also regarding ownership, ethics, regulation, and public perception. Next, we highlight the repercussions of changing data sources on statistical reporting. These include technical effects such as concept drift, bias, availability, validity, accuracy and completeness, but also the neutrality and potential discontinuation of the statistical offering. We offer a few important precautionary measures, such as enhancing robustness in both data sourcing and statistical techniques, and thorough monitoring. In doing so, machine learning-based official statistics can maintain integrity, reliability, consistency, and relevance in policy-making, decision-making, and public discourse.
Active learning for data streams: a survey
Online active learning is a paradigm in machine learning that aims to select the most informative data points to label from a data stream. The problem of minimizing the cost associated with collecting labeled observations has gained a lot of attention in recent years, particularly in real-world applications where data is only available in an unlabeled form. Annotating each observation can be time-consuming and costly, making it difficult to obtain large amounts of labeled data. To overcome this issue, many active learning strategies have been proposed in the last decades, aiming to select the most informative observations for labeling in order to improve the performance of machine learning models. These approaches can be broadly divided into two categories: static pool-based and stream-based active learning. Pool-based active learning involves selecting a subset of observations from a closed pool of unlabeled data, and it has been the focus of many surveys and literature reviews. However, the growing availability of data streams has led to an increase in the number of approaches that focus on online active learning, which involves continuously selecting and labeling observations as they arrive in a stream. This work aims to provide an overview of the most recently proposed approaches for selecting the most informative observations from data streams in real time. We review the various techniques that have been proposed and discuss their strengths and limitations, as well as the challenges and opportunities that exist in this area of research.
A Comparative Study of Self-Supervised Speech Representations in Read and Spontaneous TTS
Recent work has explored using self-supervised learning (SSL) speech representations such as wav2vec2.0 as the representation medium in standard two-stage TTS, in place of conventionally used mel-spectrograms. It is however unclear which speech SSL is the better fit for TTS, and whether or not the performance differs between read and spontaneous TTS, the later of which is arguably more challenging. This study aims at addressing these questions by testing several speech SSLs, including different layers of the same SSL, in two-stage TTS on both read and spontaneous corpora, while maintaining constant TTS model architecture and training settings. Results from listening tests show that the 9th layer of 12-layer wav2vec2.0 (ASR finetuned) outperforms other tested SSLs and mel-spectrogram, in both read and spontaneous TTS. Our work sheds light on both how speech SSL can readily improve current TTS systems, and how SSLs compare in the challenging generative task of TTS. Audio examples can be found at https://www.speech.kth.se/tts-demos/ssr_tts
Configurable EBEN: Extreme Bandwidth Extension Network to enhance body-conducted speech capture
This paper presents a configurable version of Extreme Bandwidth Extension Network (EBEN), a Generative Adversarial Network (GAN) designed to improve audio captured with body-conduction microphones. We show that although these microphones significantly reduce environmental noise, this insensitivity to ambient noise happens at the expense of the bandwidth of the speech signal acquired by the wearer of the devices. The obtained captured signals therefore require the use of signal enhancement techniques to recover the full-bandwidth speech. EBEN leverages a configurable multiband decomposition of the raw captured signal. This decomposition allows the data time domain dimensions to be reduced and the full band signal to be better controlled. The multiband representation of the captured signal is processed through a U-Net-like model, which combines feature and adversarial losses to generate an enhanced speech signal. We also benefit from this original representation in the proposed configurable discriminators architecture. The configurable EBEN approach can achieve state-of-the-art enhancement results on synthetic data with a lightweight generator that allows real-time processing.
Privacy-preserving machine learning for healthcare: open challenges and future perspectives
Machine Learning (ML) has recently shown tremendous success in modeling various healthcare prediction tasks, ranging from disease diagnosis and prognosis to patient treatment. Due to the sensitive nature of medical data, privacy must be considered along the entire ML pipeline, from model training to inference. In this paper, we conduct a review of recent literature concerning Privacy-Preserving Machine Learning (PPML) for healthcare. We primarily focus on privacy-preserving training and inference-as-a-service, and perform a comprehensive review of existing trends, identify challenges, and discuss opportunities for future research directions. The aim of this review is to guide the development of private and efficient ML models in healthcare, with the prospects of translating research efforts into real-world settings.
Physics-Inspired Interpretability Of Machine Learning Models
The ability to explain decisions made by machine learning models remains one of the most significant hurdles towards widespread adoption of AI in highly sensitive areas such as medicine, cybersecurity or autonomous driving. Great interest exists in understanding which features of the input data prompt model decision making. In this contribution, we propose a novel approach to identify relevant features of the input data, inspired by methods from the energy landscapes field, developed in the physical sciences. By identifying conserved weights within groups of minima of the loss landscapes, we can identify the drivers of model decision making. Analogues to this idea exist in the molecular sciences, where coordinate invariants or order parameters are employed to identify critical features of a molecule. However, no such approach exists for machine learning loss landscapes. We will demonstrate the applicability of energy landscape methods to machine learning models and give examples, both synthetic and from the real world, for how these methods can help to make models more interpretable.
Audio-Visual Speech Enhancement with Score-Based Generative Models
This paper introduces an audio-visual speech enhancement system that leverages score-based generative models, also known as diffusion models, conditioned on visual information. In particular, we exploit audio-visual embeddings obtained from a self-super\-vised learning model that has been fine-tuned on lipreading. The layer-wise features of its transformer-based encoder are aggregated, time-aligned, and incorporated into the noise conditional score network. Experimental evaluations show that the proposed audio-visual speech enhancement system yields improved speech quality and reduces generative artifacts such as phonetic confusions with respect to the audio-only equivalent. The latter is supported by the word error rate of a downstream automatic speech recognition model, which decreases noticeably, especially at low input signal-to-noise ratios.
Audio-visual End-to-end Multi-channel Speech Separation, Dereverberation and Recognition
Accurate recognition of cocktail party speech containing overlapping speakers, noise and reverberation remains a highly challenging task to date. Motivated by the invariance of visual modality to acoustic signal corruption, an audio-visual multi-channel speech separation, dereverberation and recognition approach featuring a full incorporation of visual information into all system components is proposed in this paper. The efficacy of the video input is consistently demonstrated in mask-based MVDR speech separation, DNN-WPE or spectral mapping (SpecM) based speech dereverberation front-end and Conformer ASR back-end. Audio-visual integrated front-end architectures performing speech separation and dereverberation in a pipelined or joint fashion via mask-based WPD are investigated. The error cost mismatch between the speech enhancement front-end and ASR back-end components is minimized by end-to-end jointly fine-tuning using either the ASR cost function alone, or its interpolation with the speech enhancement loss. Experiments were conducted on the mixture overlapped and reverberant speech data constructed using simulation or replay of the Oxford LRS2 dataset. The proposed audio-visual multi-channel speech separation, dereverberation and recognition systems consistently outperformed the comparable audio-only baseline by 9.1% and 6.2% absolute (41.7% and 36.0% relative) word error rate (WER) reductions. Consistent speech enhancement improvements were also obtained on PESQ, STOI and SRMR scores.
Incorporating Ultrasound Tongue Images for Audio-Visual Speech Enhancement
Audio-visual speech enhancement (AV-SE) aims to enhance degraded speech along with extra visual information such as lip videos, and has been shown to be more effective than audio-only speech enhancement. This paper proposes the incorporation of ultrasound tongue images to improve the performance of lip-based AV-SE systems further. To address the challenge of acquiring ultrasound tongue images during inference, we first propose to employ knowledge distillation during training to investigate the feasibility of leveraging tongue-related information without directly inputting ultrasound tongue images. Specifically, we guide an audio-lip speech enhancement student model to learn from a pre-trained audio-lip-tongue speech enhancement teacher model, thus transferring tongue-related knowledge. To better model the alignment between the lip and tongue modalities, we further propose the introduction of a lip-tongue key-value memory network into the AV-SE model. This network enables the retrieval of tongue features based on readily available lip features, thereby assisting the subsequent speech enhancement task. Experimental results demonstrate that both methods significantly improve the quality and intelligibility of the enhanced speech compared to traditional lip-based AV-SE baselines. Moreover, both proposed methods exhibit strong generalization performance on unseen speakers and in the presence of unseen noises. Furthermore, phone error rate (PER) analysis of automatic speech recognition (ASR) reveals that while all phonemes benefit from introducing ultrasound tongue images, palatal and velar consonants benefit most.
Multimodal Attention Merging for Improved Speech Recognition and Audio Event Classification
Training large foundation models using self-supervised objectives on unlabeled data, followed by fine-tuning on downstream tasks, has emerged as a standard procedure. Unfortunately, the efficacy of this approach is often constrained by both limited fine-tuning compute and scarcity in labeled downstream data. We introduce Multimodal Attention Merging (MAM), an attempt that facilitates direct knowledge transfer from attention matrices of models rooted in high resource modalities, text and images, to those in resource-constrained domains, speech and audio, employing a zero-shot paradigm. MAM reduces the relative Word Error Rate (WER) of an Automatic Speech Recognition (ASR) model by up to 6.70%, and relative classification error of an Audio Event Classification (AEC) model by 10.63%. In cases where some data/compute is available, we present Learnable-MAM, a data-driven approach to merging attention matrices, resulting in a further 2.90% relative reduction in WER for ASR and 18.42% relative reduction in AEC compared to fine-tuning.
Learning Curves for Decision Making in Supervised Machine Learning: A Survey
Learning curves are a concept from social sciences that has been adopted in the context of machine learning to assess the performance of a learning algorithm with respect to a certain resource, e.g., the number of training examples or the number of training iterations. Learning curves have important applications in several machine learning contexts, most notably in data acquisition, early stopping of model training, and model selection. For instance, learning curves can be used to model the performance of the combination of an algorithm and its hyperparameter configuration, providing insights into their potential suitability at an early stage and often expediting the algorithm selection process. Various learning curve models have been proposed to use learning curves for decision making. Some of these models answer the binary decision question of whether a given algorithm at a certain budget will outperform a certain reference performance, whereas more complex models predict the entire learning curve of an algorithm. We contribute a framework that categorises learning curve approaches using three criteria: the decision-making situation they address, the intrinsic learning curve question they answer and the type of resources they use. We survey papers from the literature and classify them into this framework.
Information Retrieval from the Digitized Books
Extracting the relevant information out of a large number of documents is a challenging and tedious task. The quality of results generated by the traditionally available full-text search engine and text-based image retrieval systems is not optimal. Information retrieval (IR) tasks become more challenging with the nontraditional language scripts, as in the case of Indic scripts. The authors have developed OCR (Optical Character Recognition) Search Engine to make an Information Retrieval & Extraction (IRE) system that replicates the current state-of-the-art methods using the IRE and Natural Language Processing (NLP) techniques. Here we have presented the study of the methods used for performing search and retrieval tasks. The details of this system, along with the statistics of the dataset (source: National Digital Library of India or NDLI), is also presented. Additionally, the ideas to further explore and add value to research in IRE are also discussed.
Learning General Audio Representations with Large-Scale Training of Patchout Audio Transformers
The success of supervised deep learning methods is largely due to their ability to learn relevant features from raw data. Deep Neural Networks (DNNs) trained on large-scale datasets are capable of capturing a diverse set of features, and learning a representation that can generalize onto unseen tasks and datasets that are from the same domain. Hence, these models can be used as powerful feature extractors, in combination with shallower models as classifiers, for smaller tasks and datasets where the amount of training data is insufficient for learning an end-to-end model from scratch. During the past years, Convolutional Neural Networks (CNNs) have largely been the method of choice for audio processing. However, recently attention-based transformer models have demonstrated great potential in supervised settings, outperforming CNNs. In this work, we investigate the use of audio transformers trained on large-scale datasets to learn general-purpose representations. We study how the different setups in these audio transformers affect the quality of their embeddings. We experiment with the models' time resolution, extracted embedding level, and receptive fields in order to see how they affect performance on a variety of tasks and datasets, following the HEAR 2021 NeurIPS challenge evaluation setup. Our results show that representations extracted by audio transformers outperform CNN representations. Furthermore, we will show that transformers trained on Audioset can be extremely effective representation extractors for a wide range of downstream tasks.
Pretrained audio neural networks for Speech emotion recognition in Portuguese
The goal of speech emotion recognition (SER) is to identify the emotional aspects of speech. The SER challenge for Brazilian Portuguese speech was proposed with short snippets of Portuguese which are classified as neutral, non-neutral female and non-neutral male according to paralinguistic elements (laughing, crying, etc). This dataset contains about $50$ minutes of Brazilian Portuguese speech. As the dataset leans on the small side, we investigate whether a combination of transfer learning and data augmentation techniques can produce positive results. Thus, by combining a data augmentation technique called SpecAugment, with the use of Pretrained Audio Neural Networks (PANNs) for transfer learning we are able to obtain interesting results. The PANNs (CNN6, CNN10 and CNN14) are pretrained on a large dataset called AudioSet containing more than $5000$ hours of audio. They were finetuned on the SER dataset and the best performing model (CNN10) on the validation set was submitted to the challenge, achieving an $F1$ score of $0.73$ up from $0.54$ from the baselines provided by the challenge. Moreover, we also tested the use of Transformer neural architecture, pretrained on about $600$ hours of Brazilian Portuguese audio data. Transformers, as well as more complex models of PANNs (CNN14), fail to generalize to the test set in the SER dataset and do not beat the baseline. Considering the limitation of the dataset sizes, currently the best approach for SER is using PANNs (specifically, CNN6 and CNN10).
DeLoRes: Decorrelating Latent Spaces for Low-Resource Audio Representation Learning
Inspired by the recent progress in self-supervised learning for computer vision, in this paper we introduce DeLoRes, a new general-purpose audio representation learning approach. Our main objective is to make our network learn representations in a resource-constrained setting (both data and compute), that can generalize well across a diverse set of downstream tasks. Inspired from the Barlow Twins objective function, we propose to learn embeddings that are invariant to distortions of an input audio sample, while making sure that they contain non-redundant information about the sample. To achieve this, we measure the cross-correlation matrix between the outputs of two identical networks fed with distorted versions of an audio segment sampled from an audio file and make it as close to the identity matrix as possible. We use a combination of a small subset of the large-scale AudioSet dataset and FSD50K for self-supervised learning and are able to learn with less than half the parameters compared to state-of-the-art algorithms. For evaluation, we transfer these learned representations to 9 downstream classification tasks, including speech, music, and animal sounds, and show competitive results under different evaluation setups. In addition to being simple and intuitive, our pre-training algorithm is amenable to compute through its inherent nature of construction and does not require careful implementation details to avoid trivial or degenerate solutions. Furthermore, we conduct ablation studies on our results and make all our code and pre-trained models publicly available https://github.com/Speech-Lab-IITM/DeLoRes.
Supervised Contrastive Learning for Accented Speech Recognition
Neural network based speech recognition systems suffer from performance degradation due to accented speech, especially unfamiliar accents. In this paper, we study the supervised contrastive learning framework for accented speech recognition. To build different views (similar "positive" data samples) for contrastive learning, three data augmentation techniques including noise injection, spectrogram augmentation and TTS-same-sentence generation are further investigated. From the experiments on the Common Voice dataset, we have shown that contrastive learning helps to build data-augmentation invariant and pronunciation invariant representations, which significantly outperforms traditional joint training methods in both zero-shot and full-shot settings. Experiments show that contrastive learning can improve accuracy by 3.66% (zero-shot) and 3.78% (full-shot) on average, comparing to the joint training method.
Meta-TTS: Meta-Learning for Few-Shot Speaker Adaptive Text-to-Speech
Personalizing a speech synthesis system is a highly desired application, where the system can generate speech with the user's voice with rare enrolled recordings. There are two main approaches to build such a system in recent works: speaker adaptation and speaker encoding. On the one hand, speaker adaptation methods fine-tune a trained multi-speaker text-to-speech (TTS) model with few enrolled samples. However, they require at least thousands of fine-tuning steps for high-quality adaptation, making it hard to apply on devices. On the other hand, speaker encoding methods encode enrollment utterances into a speaker embedding. The trained TTS model can synthesize the user's speech conditioned on the corresponding speaker embedding. Nevertheless, the speaker encoder suffers from the generalization gap between the seen and unseen speakers. In this paper, we propose applying a meta-learning algorithm to the speaker adaptation method. More specifically, we use Model Agnostic Meta-Learning (MAML) as the training algorithm of a multi-speaker TTS model, which aims to find a great meta-initialization to adapt the model to any few-shot speaker adaptation tasks quickly. Therefore, we can also adapt the meta-trained TTS model to unseen speakers efficiently. Our experiments compare the proposed method (Meta-TTS) with two baselines: a speaker adaptation method baseline and a speaker encoding method baseline. The evaluation results show that Meta-TTS can synthesize high speaker-similarity speech from few enrollment samples with fewer adaptation steps than the speaker adaptation baseline and outperforms the speaker encoding baseline under the same training scheme. When the speaker encoder of the baseline is pre-trained with extra 8371 speakers of data, Meta-TTS can still outperform the baseline on LibriTTS dataset and achieve comparable results on VCTK dataset.
Comparison of Self-Supervised Speech Pre-Training Methods on Flemish Dutch
Recent research in speech processing exhibits a growing interest in unsupervised and self-supervised representation learning from unlabelled data to alleviate the need for large amounts of annotated data. We investigate several popular pre-training methods and apply them to Flemish Dutch. We compare off-the-shelf English pre-trained models to models trained on an increasing amount of Flemish data. We find that the most important factors for positive transfer to downstream speech recognition tasks include a substantial amount of data and a matching pre-training domain. Ideally, we also finetune on an annotated subset in the target language. All pre-trained models improve linear phone separability in Flemish, but not all methods improve Automatic Speech Recognition. We experience superior performance with wav2vec 2.0 and we obtain a 30% WER improvement by finetuning the multilingually pre-trained XLSR-53 model on Flemish Dutch, after integration into an HMM-DNN acoustic model.
Learning Audio-Visual Dereverberation
Reverberation not only degrades the quality of speech for human perception, but also severely impacts the accuracy of automatic speech recognition. Prior work attempts to remove reverberation based on the audio modality only. Our idea is to learn to dereverberate speech from audio-visual observations. The visual environment surrounding a human speaker reveals important cues about the room geometry, materials, and speaker location, all of which influence the precise reverberation effects. We introduce Visually-Informed Dereverberation of Audio (VIDA), an end-to-end approach that learns to remove reverberation based on both the observed monaural sound and visual scene. In support of this new task, we develop a large-scale dataset SoundSpaces-Speech that uses realistic acoustic renderings of speech in real-world 3D scans of homes offering a variety of room acoustics. Demonstrating our approach on both simulated and real imagery for speech enhancement, speech recognition, and speaker identification, we show it achieves state-of-the-art performance and substantially improves over audio-only methods.
COALA: Co-Aligned Autoencoders for Learning Semantically Enriched Audio Representations
Audio representation learning based on deep neural networks (DNNs) emerged as an alternative approach to hand-crafted features. For achieving high performance, DNNs often need a large amount of annotated data which can be difficult and costly to obtain. In this paper, we propose a method for learning audio representations, aligning the learned latent representations of audio and associated tags. Aligning is done by maximizing the agreement of the latent representations of audio and tags, using a contrastive loss. The result is an audio embedding model which reflects acoustic and semantic characteristics of sounds. We evaluate the quality of our embedding model, measuring its performance as a feature extractor on three different tasks (namely, sound event recognition, and music genre and musical instrument classification), and investigate what type of characteristics the model captures. Our results are promising, sometimes in par with the state-of-the-art in the considered tasks and the embeddings produced with our method are well correlated with some acoustic descriptors.
Interpretable Representation Learning for Speech and Audio Signals Based on Relevance Weighting
The learning of interpretable representations from raw data presents significant challenges for time series data like speech. In this work, we propose a relevance weighting scheme that allows the interpretation of the speech representations during the forward propagation of the model itself. The relevance weighting is achieved using a sub-network approach that performs the task of feature selection. A relevance sub-network, applied on the output of first layer of a convolutional neural network model operating on raw speech signals, acts as an acoustic filterbank (FB) layer with relevance weighting. A similar relevance sub-network applied on the second convolutional layer performs modulation filterbank learning with relevance weighting. The full acoustic model consisting of relevance sub-networks, convolutional layers and feed-forward layers is trained for a speech recognition task on noisy and reverberant speech in the Aurora-4, CHiME-3 and VOiCES datasets. The proposed representation learning framework is also applied for the task of sound classification in the UrbanSound8K dataset. A detailed analysis of the relevance weights learned by the model reveals that the relevance weights capture information regarding the underlying speech/audio content. In addition, speech recognition and sound classification experiments reveal that the incorporation of relevance weighting in the neural network architecture improves the performance significantly.
Learning Speech Representations from Raw Audio by Joint Audiovisual Self-Supervision
The intuitive interaction between the audio and visual modalities is valuable for cross-modal self-supervised learning. This concept has been demonstrated for generic audiovisual tasks like video action recognition and acoustic scene classification. However, self-supervision remains under-explored for audiovisual speech. We propose a method to learn self-supervised speech representations from the raw audio waveform. We train a raw audio encoder by combining audio-only self-supervision (by predicting informative audio attributes) with visual self-supervision (by generating talking faces from audio). The visual pretext task drives the audio representations to capture information related to lip movements. This enriches the audio encoder with visual information and the encoder can be used for evaluation without the visual modality. Our method attains competitive performance with respect to existing self-supervised audio features on established isolated word classification benchmarks, and significantly outperforms other methods at learning from fewer labels. Notably, our method also outperforms fully supervised training, thus providing a strong initialization for speech related tasks. Our results demonstrate the potential of multimodal self-supervision in audiovisual speech for learning good audio representations.
DOME: Recommendations for supervised machine learning validation in biology
Modern biology frequently relies on machine learning to provide predictions and improve decision processes. There have been recent calls for more scrutiny on machine learning performance and possible limitations. Here we present a set of community-wide recommendations aiming to help establish standards of supervised machine learning validation in biology. Adopting a structured methods description for machine learning based on data, optimization, model, evaluation (DOME) will aim to help both reviewers and readers to better understand and assess the performance and limitations of a method or outcome. The recommendations are formulated as questions to anyone wishing to pursue implementation of a machine learning algorithm. Answers to these questions can be easily included in the supplementary material of published papers.
How to Teach DNNs to Pay Attention to the Visual Modality in Speech Recognition
Audio-Visual Speech Recognition (AVSR) seeks to model, and thereby exploit, the dynamic relationship between a human voice and the corresponding mouth movements. A recently proposed multimodal fusion strategy, AV Align, based on state-of-the-art sequence to sequence neural networks, attempts to model this relationship by explicitly aligning the acoustic and visual representations of speech. This study investigates the inner workings of AV Align and visualises the audio-visual alignment patterns. Our experiments are performed on two of the largest publicly available AVSR datasets, TCD-TIMIT and LRS2. We find that AV Align learns to align acoustic and visual representations of speech at the frame level on TCD-TIMIT in a generally monotonic pattern. We also determine the cause of initially seeing no improvement over audio-only speech recognition on the more challenging LRS2. We propose a regularisation method which involves predicting lip-related Action Units from visual representations. Our regularisation method leads to better exploitation of the visual modality, with performance improvements between 7% and 30% depending on the noise level. Furthermore, we show that the alternative Watch, Listen, Attend, and Spell network is affected by the same problem as AV Align, and that our proposed approach can effectively help it learn visual representations. Our findings validate the suitability of the regularisation method to AVSR and encourage researchers to rethink the multimodal convergence problem when having one dominant modality.
Recurrent Neural Network Transducer for Audio-Visual Speech Recognition
This work presents a large-scale audio-visual speech recognition system based on a recurrent neural network transducer (RNN-T) architecture. To support the development of such a system, we built a large audio-visual (A/V) dataset of segmented utterances extracted from YouTube public videos, leading to 31k hours of audio-visual training content. The performance of an audio-only, visual-only, and audio-visual system are compared on two large-vocabulary test sets: a set of utterance segments from public YouTube videos called YTDEV18 and the publicly available LRS3-TED set. To highlight the contribution of the visual modality, we also evaluated the performance of our system on the YTDEV18 set artificially corrupted with background noise and overlapping speech. To the best of our knowledge, our system significantly improves the state-of-the-art on the LRS3-TED set.
Audio-visual Speech Enhancement Using Conditional Variational Auto-Encoders
Variational auto-encoders (VAEs) are deep generative latent variable models that can be used for learning the distribution of complex data. VAEs have been successfully used to learn a probabilistic prior over speech signals, which is then used to perform speech enhancement. One advantage of this generative approach is that it does not require pairs of clean and noisy speech signals at training. In this paper, we propose audio-visual variants of VAEs for single-channel and speaker-independent speech enhancement. We develop a conditional VAE (CVAE) where the audio speech generative process is conditioned on visual information of the lip region. At test time, the audio-visual speech generative model is combined with a noise model based on nonnegative matrix factorization, and speech enhancement relies on a Monte Carlo expectation-maximization algorithm. Experiments are conducted with the recently published NTCD-TIMIT dataset as well as the GRID corpus. The results confirm that the proposed audio-visual CVAE effectively fuses audio and visual information, and it improves the speech enhancement performance compared with the audio-only VAE model, especially when the speech signal is highly corrupted by noise. We also show that the proposed unsupervised audio-visual speech enhancement approach outperforms a state-of-the-art supervised deep learning method.
Leveraging End-to-End Speech Recognition with Neural Architecture Search
Deep neural networks (DNNs) have been demonstrated to outperform many traditional machine learning algorithms in Automatic Speech Recognition (ASR). In this paper, we show that a large improvement in the accuracy of deep speech models can be achieved with effective Neural Architecture Optimization at a very low computational cost. Phone recognition tests with the popular LibriSpeech and TIMIT benchmarks proved this fact by displaying the ability to discover and train novel candidate models within a few hours (less than a day) many times faster than the attention-based seq2seq models. Our method achieves test error of 7% Word Error Rate (WER) on the LibriSpeech corpus and 13% Phone Error Rate (PER) on the TIMIT corpus, on par with state-of-the-art results.