Abstract:Curriculum learning (CL) describes a machine learning training strategy in which samples are gradually introduced into the training process based on their difficulty. Despite a partially contradictory body of evidence in the literature, CL finds popularity in deep learning research due to its promise of leveraging human-inspired curricula to achieve higher model performance. Yet, the subjectivity and biases that follow any necessary definition of difficulty, especially for those found in orderings derived from models or training statistics, have rarely been investigated. To shed more light on the underlying unanswered questions, we conduct an extensive study on the robustness and similarity of the most common scoring functions for sample difficulty estimation, as well as their potential benefits in CL, using the popular benchmark dataset CIFAR-10 and the acoustic scene classification task from the DCASE2020 challenge as representatives of computer vision and computer audition, respectively. We report a strong dependence of scoring functions on the training setting, including randomness, which can partly be mitigated through ensemble scoring. While we do not find a general advantage of CL over uniform sampling, we observe that the ordering in which data is presented for CL-based training plays an important role in model performance. Furthermore, we find that the robustness of scoring functions across random seeds positively correlates with CL performance. Finally, we uncover that models trained with different CL strategies complement each other by boosting predictive power through late fusion, likely due to differences in the learnt concepts. Alongside our findings, we release the aucurriculum toolkit (https://github.com/autrainer/aucurriculum), implementing sample difficulty and CL-based training in a modular fashion.
Abstract:While current emotional text-to-speech (TTS) systems can generate highly intelligible emotional speech, achieving fine control over emotion rendering of the output speech still remains a significant challenge. In this paper, we introduce ParaEVITS, a novel emotional TTS framework that leverages the compositionality of natural language to enhance control over emotional rendering. By incorporating a text-audio encoder inspired by ParaCLAP, a contrastive language-audio pretraining (CLAP) model for computational paralinguistics, the diffusion model is trained to generate emotional embeddings based on textual emotional style descriptions. Our framework first trains on reference audio using the audio encoder, then fine-tunes a diffusion model to process textual inputs from ParaCLAP's text encoder. During inference, speech attributes such as pitch, jitter, and loudness are manipulated using only textual conditioning. Our experiments demonstrate that ParaEVITS effectively control emotion rendering without compromising speech quality. Speech demos are publicly available.
Abstract:Neural network models for audio tasks, such as automatic speech recognition (ASR) and acoustic scene classification (ASC), are susceptible to noise contamination for real-life applications. To improve audio quality, an enhancement module, which can be developed independently, is explicitly used at the front-end of the target audio applications. In this paper, we present an end-to-end learning solution to jointly optimise the models for audio enhancement (AE) and the subsequent applications. To guide the optimisation of the AE module towards a target application, and especially to overcome difficult samples, we make use of the sample-wise performance measure as an indication of sample importance. In experiments, we consider four representative applications to evaluate our training paradigm, i.e., ASR, speech command recognition (SCR), speech emotion recognition (SER), and ASC. These applications are associated with speech and non-speech tasks concerning semantic and non-semantic features, transient and global information, and the experimental results indicate that our proposed approach can considerably boost the noise robustness of the models, especially at low signal-to-noise ratios (SNRs), for a wide range of computer audition tasks in everyday-life noisy environments.
Abstract:Abusive content in online social networks is a well-known problem that can cause serious psychological harm and incite hatred. The ability to upload audio data increases the importance of developing methods to detect abusive content in speech recordings. However, simply transferring the mechanisms from written abuse detection would ignore relevant information such as emotion and tone. In addition, many current algorithms require training in the specific language for which they are being used. This paper proposes to use acoustic and prosodic features to classify abusive content. We used the ADIMA data set, which contains recordings from ten Indic languages, and trained different models in multilingual and cross-lingual settings. Our results show that it is possible to classify abusive and non-abusive content using only acoustic and prosodic features. The most important and influential features are discussed.
Abstract:Foundation models (FMs) are increasingly spearheading recent advances on a variety of tasks that fall under the purview of computer audition -- the use of machines to understand sounds. They feature several advantages over traditional pipelines: among others, the ability to consolidate multiple tasks in a single model, the option to leverage knowledge from other modalities, and the readily-available interaction with human users. Naturally, these promises have created substantial excitement in the audio community, and have led to a wave of early attempts to build new, general-purpose foundation models for audio. In the present contribution, we give an overview of computational audio analysis as it transitions from traditional pipelines towards auditory foundation models. Our work highlights the key operating principles that underpin those models, and showcases how they can accommodate multiple tasks that the audio community previously tackled separately.
Abstract:Contrastive language-audio pretraining (CLAP) has recently emerged as a method for making audio analysis more generalisable. Specifically, CLAP-style models are able to `answer' a diverse set of language queries, extending the capabilities of audio models beyond a closed set of labels. However, CLAP relies on a large set of (audio, query) pairs for pretraining. While such sets are available for general audio tasks, like captioning or sound event detection, there are no datasets with matched audio and text queries for computational paralinguistic (CP) tasks. As a result, the community relies on generic CLAP models trained for general audio with limited success. In the present study, we explore training considerations for ParaCLAP, a CLAP-style model suited to CP, including a novel process for creating audio-language queries. We demonstrate its effectiveness on a set of computational paralinguistic tasks, where it is shown to surpass the performance of open-source state-of-the-art models.
Abstract:We revisit the INTERSPEECH 2009 Emotion Challenge -- the first ever speech emotion recognition (SER) challenge -- and evaluate a series of deep learning models that are representative of the major advances in SER research in the time since then. We start by training each model using a fixed set of hyperparameters, and further fine-tune the best-performing models of that initial setup with a grid search. Results are always reported on the official test set with a separate validation set only used for early stopping. Most models score below or close to the official baseline, while they marginally outperform the original challenge winners after hyperparameter tuning. Our work illustrates that, despite recent progress, FAU-AIBO remains a very challenging benchmark. An interesting corollary is that newer methods do not consistently outperform older ones, showing that progress towards `solving' SER is not necessarily monotonic.
Abstract:Chronic obstructive pulmonary disease (COPD) is a serious inflammatory lung disease affecting millions of people around the world. Due to an obstructed airflow from the lungs, it also becomes manifest in patients' vocal behaviour. Of particular importance is the detection of an exacerbation episode, which marks an acute phase and often requires hospitalisation and treatment. Previous work has shown that it is possible to distinguish between a pre- and a post-treatment state using automatic analysis of read speech. In this contribution, we examine whether sustained vowels can provide a complementary lens for telling apart these two states. Using a cohort of 50 patients, we show that the inclusion of sustained vowels can improve performance to up to 79\% unweighted average recall, from a 71\% baseline using read speech. We further identify and interpret the most important acoustic features that characterise the manifestation of COPD in sustained vowels.
Abstract:In recent decades, running has become an increasingly popular pastime activity due to its accessibility, ease of practice, and anticipated health benefits. However, the risk of running-related injuries is substantial for runners of different experience levels. Several common forms of injuries result from overuse -- extending beyond the recommended running time and intensity. Recently, audio-based tracking has emerged as yet another modality for monitoring running behaviour and performance, with previous studies largely concentrating on predicting runner fatigue. In this work, we investigate audio-based step count estimation during outdoor running, achieving a mean absolute error of 1.098 in window-based step-count differences and a Pearson correlation coefficient of 0.479 when predicting the number of steps in a 5-second window of audio. Our work thus showcases the feasibility of audio-based monitoring for estimating important physiological variables and lays the foundations for further utilising audio sensors for a more thorough characterisation of runner behaviour.
Abstract:The expression of emotion is highly individualistic. However, contemporary speech emotion recognition (SER) systems typically rely on population-level models that adopt a `one-size-fits-all' approach for predicting emotion. Moreover, standard evaluation practices measure performance also on the population level, thus failing to characterise how models work across different speakers. In the present contribution, we present a new method for capitalising on individual differences to adapt an SER model to each new speaker using a minimal set of enrolment utterances. In addition, we present novel evaluation schemes for measuring fairness across different speakers. Our findings show that aggregated evaluation metrics may obfuscate fairness issues on the individual-level, which are uncovered by our evaluation, and that our proposed method can improve performance both in aggregated and disaggregated terms.