Nonlinear functional regression by functional deep neural network with kernel embedding

With the rapid development of deep learning in various fields of science and technology, such as speech recognition, image classification, and natural language processing, recently it is also widely applied in the functional data analysis (FDA) with some empirical success. However, due to the infinite dimensional input, we need a powerful dimension reduction method for functional learning tasks, especially for the nonlinear functional regression. In this paper, based on the idea of smooth kernel integral transformation, we propose a functional deep neural network with an efficient and fully data-dependent dimension reduction method. The architecture of our functional net consists of a kernel embedding step: an integral transformation with a data-dependent smooth kernel; a projection step: a dimension reduction by projection with eigenfunction basis based on the embedding kernel; and finally an expressive deep ReLU neural network for the prediction. The utilization of smooth kernel embedding enables our functional net to be discretization invariant, efficient, and robust to noisy observations, capable of utilizing information in both input functions and responses data, and have a low requirement on the number of discrete points for an unimpaired generalization performance. We conduct theoretical analysis including approximation error and generalization error analysis, and numerical simulations to verify these advantages of our functional net.

A comparative analysis between Conformer-Transducer, Whisper, and wav2vec2 for improving the child speech recognition

Automatic Speech Recognition (ASR) systems have progressed significantly in their performance on adult speech data; however, transcribing child speech remains challenging due to the acoustic differences in the characteristics of child and adult voices. This work aims to explore the potential of adapting state-of-the-art Conformer-transducer models to child speech to improve child speech recognition performance. Furthermore, the results are compared with those of self-supervised wav2vec2 models and semi-supervised multi-domain Whisper models that were previously finetuned on the same data. We demonstrate that finetuning Conformer-transducer models on child speech yields significant improvements in ASR performance on child speech, compared to the non-finetuned models. We also show Whisper and wav2vec2 adaptation on different child speech datasets. Our detailed comparative analysis shows that wav2vec2 provides the most consistent performance improvements among the three methods studied.

Whisper in Focus: Enhancing Stuttered Speech Classification with Encoder Layer Optimization

In recent years, advancements in the field of speech processing have led to cutting-edge deep learning algorithms with immense potential for real-world applications. The automated identification of stuttered speech is one of such applications that the researchers are addressing by employing deep learning techniques. Recently, researchers have utilized Wav2vec2.0, a speech recognition model to classify disfluency types in stuttered speech. Although Wav2vec2.0 has shown commendable results, its ability to generalize across all disfluency types is limited. In addition, since its base model uses 12 encoder layers, it is considered a resource-intensive model. Our study unravels the capabilities of Whisper for the classification of disfluency types in stuttered speech. We have made notable contributions in three pivotal areas: enhancing the quality of SEP28-k benchmark dataset, exploration of Whisper for classification, and introducing an efficient encoder layer freezing strategy. The optimized Whisper model has achieved the average F1-score of 0.81, which proffers its abilities. This study also unwinds the significance of deeper encoder layers in the identification of disfluency types, as the results demonstrate their greater contribution compared to initial layers. This research represents substantial contributions, shifting the emphasis towards an efficient solution, thereby thriving towards prospective innovation.

Selective HuBERT: Self-Supervised Pre-Training for Target Speaker in Clean and Mixture Speech

Self-supervised pre-trained speech models were shown effective for various downstream speech processing tasks. Since they are mainly pre-trained to map input speech to pseudo-labels, the resulting representations are only effective for the type of pre-train data used, either clean or mixture speech. With the idea of selective auditory attention, we propose a novel pre-training solution called Selective-HuBERT, or SHuBERT, which learns the selective extraction of target speech representations from either clean or mixture speech. Specifically, SHuBERT is trained to predict pseudo labels of a target speaker, conditioned on an enrolled speech from the target speaker. By doing so, SHuBERT is expected to selectively attend to the target speaker in a complex acoustic environment, thus benefiting various downstream tasks. We further introduce a dual-path training strategy and use the cross-correlation constraint between the two branches to encourage the model to generate noise-invariant representation. Experiments on SUPERB benchmark and LibriMix dataset demonstrate the universality and noise-robustness of SHuBERT. Furthermore, we find that our high-quality representation can be easily integrated with conventional supervised learning methods to achieve significant performance, even under extremely low-resource labeled data.

Fine-tuning convergence model in Bengali speech recognition

Research on speech recognition has attracted considerable interest due to the difficult task of segmenting uninterrupted speech. Among various languages, Bengali features distinct rhythmic patterns and tones, making it particularly difficult to recognize and lacking an efficient commercial recognition method. In order to improve the automatic speech recognition model for Bengali, our team has chosen to utilize the wave2vec 2.0 pre-trained model, which has undergone convergence for fine-tuning. Regarding Word Error Rate (WER), the learning rate and dropout parameters were fine-tuned, and after the model training was stable, attempts were made to enlarge the training set ratio, which improved the model's performance. Consequently, there was a notable enhancement in the WER from 0.508 to 0.437 on the test set of the publicly listed official dataset. Afterwards, the training and validation sets were merged, creating a comprehensive dataset that was used as the training set, achieving a remarkable WER of 0.436.

Enhancing Zero-Shot Multi-Speaker TTS with Negated Speaker Representations

Zero-shot multi-speaker TTS aims to synthesize speech with the voice of a chosen target speaker without any fine-tuning. Prevailing methods, however, encounter limitations at adapting to new speakers of out-of-domain settings, primarily due to inadequate speaker disentanglement and content leakage. To overcome these constraints, we propose an innovative negation feature learning paradigm that models decoupled speaker attributes as deviations from the complete audio representation by utilizing the subtraction operation. By eliminating superfluous content information from the speaker representation, our negation scheme not only mitigates content leakage, thereby enhancing synthesis robustness, but also improves speaker fidelity. In addition, to facilitate the learning of diverse speaker attributes, we leverage multi-stream Transformers, which retain multiple hypotheses and instigate a training paradigm akin to ensemble learning. To unify these hypotheses and realize the final speaker representation, we employ attention pooling. Finally, in light of the imperative to generate target text utterances in the desired voice, we adopt adaptive layer normalizations to effectively fuse the previously generated speaker representation with the target text representations, as opposed to mere concatenation of the text and audio modalities. Extensive experiments and validations substantiate the efficacy of our proposed approach in preserving and harnessing speaker-specific attributes vis-`a-vis alternative baseline models.

Multimodal Data and Resource Efficient Device-Directed Speech Detection with Large Foundation Models

Interactions with virtual assistants typically start with a trigger phrase followed by a command. In this work, we explore the possibility of making these interactions more natural by eliminating the need for a trigger phrase. Our goal is to determine whether a user addressed the virtual assistant based on signals obtained from the streaming audio recorded by the device microphone. We address this task by combining 1-best hypotheses and decoder signals from an automatic speech recognition system with acoustic representations from an audio encoder as input features to a large language model (LLM). In particular, we are interested in data and resource efficient systems that require only a small amount of training data and can operate in scenarios with only a single frozen LLM available on a device. For this reason, our model is trained on 80k or less examples of multimodal data using a combination of low-rank adaptation and prefix tuning. We compare the proposed system to unimodal baselines and show that the multimodal approach achieves lower equal-error-rates (EERs), while using only a fraction of the training data. We also show that low-dimensional specialized audio representations lead to lower EERs than high-dimensional general audio representations.

FastInject: Injecting Unpaired Text Data into CTC-based ASR training

Recently, connectionist temporal classification (CTC)-based end-to-end (E2E) automatic speech recognition (ASR) models have achieved impressive results, especially with the development of self-supervised learning. However, E2E ASR models trained on paired speech-text data often suffer from domain shifts from training to testing. To alleviate this issue, this paper proposes a flat-start joint training method, named FastInject, which efficiently injects multi-domain unpaired text data into CTC-based ASR training. To maintain training efficiency, text units are pre-upsampled, and their representations are fed into the CTC model along with speech features. To bridge the modality gap between speech and text, an attention-based modality matching mechanism (AM3) is proposed, which retains the E2E flat-start training. Experiments show that the proposed FastInject gave a 22\% relative WER reduction (WERR) for intra-domain Librispeech-100h data and 20\% relative WERR on out-of-domain test sets.

TSST: A Benchmark and Evaluation Models for Text Speech-Style Transfer

Text style is highly abstract, as it encompasses various aspects of a speaker's characteristics, habits, logical thinking, and the content they express. However, previous text-style transfer tasks have primarily focused on data-driven approaches, lacking in-depth analysis and research from the perspectives of linguistics and cognitive science. In this paper, we introduce a novel task called Text Speech-Style Transfer (TSST). The main objective is to further explore topics related to human cognition, such as personality and emotion, based on the capabilities of existing LLMs. Considering the objective of our task and the distinctive characteristics of oral speech in real-life scenarios, we trained multi-dimension (i.e. filler words, vividness, interactivity, emotionality) evaluation models for the TSST and validated their correlation with human assessments. We thoroughly analyze the performance of several large language models (LLMs) and identify areas where further improvement is needed. Moreover, driven by our evaluation models, we have released a new corpus that improves the capabilities of LLMs in generating text with speech-style characteristics. In summary, we present the TSST task, a new benchmark for style transfer and emphasizing human-oriented evaluation, exploring and advancing the performance of current LLMs.

FedCPC: An Effective Federated Contrastive Learning Method for Privacy Preserving Early-Stage Alzheimer's Speech Detection

The early-stage Alzheimer's disease (AD) detection has been considered an important field of medical studies. Like traditional machine learning methods, speech-based automatic detection also suffers from data privacy risks because the data of specific patients are exclusive to each medical institution. A common practice is to use federated learning to protect the patients' data privacy. However, its distributed learning process also causes performance reduction. To alleviate this problem while protecting user privacy, we propose a federated contrastive pre-training (FedCPC) performed before federated training for AD speech detection, which can learn a better representation from raw data and enables different clients to share data in the pre-training and training stages. Experimental results demonstrate that the proposed methods can achieve satisfactory performance while preserving data privacy.