On the Relation Between Speech Quality and Quantized Latent Representations of Neural Codecs

Neural audio signal codecs have attracted significant attention in recent years. In essence, the impressive low bitrate achieved by such encoders is enabled by learning an abstract representation that captures the properties of encoded signals, e.g., speech. In this work, we investigate the relation between the latent representation of the input signal learned by a neural codec and the quality of speech signals. To do so, we introduce Latent-representation-to-Quantization error Ratio (LQR) measures, which quantify the distance from the idealized neural codec's speech signal model for a given speech signal. We compare the proposed metrics to intrusive measures as well as data-driven supervised methods using two subjective speech quality datasets. This analysis shows that the proposed LQR correlates strongly (up to 0.9 Pearson's correlation) with the subjective quality of speech. Despite being a non-intrusive metric, this yields a competitive performance with, or even better than, other pre-trained and intrusive measures. These results show that LQR is a promising basis for more sophisticated speech quality measures.

Transparent NLP: Using RAG and LLM Alignment for Privacy Q&A

The transparency principle of the General Data Protection Regulation (GDPR) requires data processing information to be clear, precise, and accessible. While language models show promise in this context, their probabilistic nature complicates truthfulness and comprehensibility. This paper examines state-of-the-art Retrieval Augmented Generation (RAG) systems enhanced with alignment techniques to fulfill GDPR obligations. We evaluate RAG systems incorporating an alignment module like Rewindable Auto-regressive Inference (RAIN) and our proposed multidimensional extension, MultiRAIN, using a Privacy Q&A dataset. Responses are optimized for preciseness and comprehensibility and are assessed through 21 metrics, including deterministic and large language model-based evaluations. Our results show that RAG systems with an alignment module outperform baseline RAG systems on most metrics, though none fully match human answers. Principal component analysis of the results reveals complex interactions between metrics, highlighting the need to refine metrics. This study provides a foundation for integrating advanced natural language processing systems into legal compliance frameworks.

Align-ULCNet: Towards Low-Complexity and Robust Acoustic Echo and Noise Reduction

The successful deployment of deep learning-based acoustic echo and noise reduction (AENR) methods in consumer devices has spurred interest in developing low-complexity solutions, while emphasizing the need for robust performance in real-life applications. In this work, we propose a hybrid approach to enhance the state-of-the-art (SOTA) ULCNet model by integrating time alignment and parallel encoder blocks for the model inputs, resulting in better echo reduction and comparable noise reduction performance to existing SOTA methods. We also propose a channel-wise sampling-based feature reorientation method, ensuring robust performance across many challenging scenarios, while maintaining overall low computational and memory requirements.

GAN-Based Speech Enhancement for Low SNR Using Latent Feature Conditioning

Enhancing speech quality under adverse SNR conditions remains a significant challenge for discriminative deep neural network (DNN)-based approaches. In this work, we propose DisCoGAN, which is a time-frequency-domain generative adversarial network (GAN) conditioned by the latent features of a discriminative model pre-trained for speech enhancement in low SNR scenarios. Our proposed method achieves superior performance compared to state-of-the-arts discriminative methods and also surpasses end-to-end (E2E) trained GAN models. We also investigate the impact of various configurations for conditioning the proposed GAN model with the discriminative model and assess their influence on enhancing speech quality

A Hybrid Approach for Low-Complexity Joint Acoustic Echo and Noise Reduction

Deep learning-based methods that jointly perform the task of acoustic echo and noise reduction (AENR) often require high memory and computational resources, making them unsuitable for real-time deployment on low-resource platforms such as embedded devices. We propose a low-complexity hybrid approach for joint AENR by employing a single model to suppress both residual echo and noise components. Specifically, we integrate the state-of-the-art (SOTA) ULCNet model, which was originally proposed to achieve ultra-low complexity noise suppression, in a hybrid system and train it for joint AENR. We show that the proposed approach achieves better echo reduction and comparable noise reduction performance with much lower computational complexity and memory requirements than all considered SOTA methods, at the cost of slight degradation in speech quality.

Comparative Analysis Of Discriminative Deep Learning-Based Noise Reduction Methods In Low SNR Scenarios

In this study, we conduct a comparative analysis of deep learning-based noise reduction methods in low signal-to-noise ratio (SNR) scenarios. Our investigation primarily focuses on five key aspects: The impact of training data, the influence of various loss functions, the effectiveness of direct and indirect speech estimation techniques, the efficacy of masking, mapping, and deep filtering methodologies, and the exploration of different model capacities on noise reduction performance and speech quality. Through comprehensive experimentation, we provide insights into the strengths, weaknesses, and applicability of these methods in low SNR environments. The findings derived from our analysis are intended to assist both researchers and practitioners in selecting better techniques tailored to their specific applications within the domain of low SNR noise reduction.

Blind Acoustic Parameter Estimation Through Task-Agnostic Embeddings Using Latent Approximations

We present a method for blind acoustic parameter estimation from single-channel reverberant speech. The method is structured into three stages. In the first stage, a variational auto-encoder is trained to extract latent representations of acoustic impulse responses represented as mel-spectrograms. In the second stage, a separate speech encoder is trained to estimate low-dimensional representations from short segments of reverberant speech. Finally, the pre-trained speech encoder is combined with a small regression model and evaluated on two parameter regression tasks. Experimentally, the proposed method is shown to outperform a fully end-to-end trained baseline model.

Meta Learning Text-to-Speech Synthesis in over 7000 Languages

In this work, we take on the challenging task of building a single text-to-speech synthesis system that is capable of generating speech in over 7000 languages, many of which lack sufficient data for traditional TTS development. By leveraging a novel integration of massively multilingual pretraining and meta learning to approximate language representations, our approach enables zero-shot speech synthesis in languages without any available data. We validate our system's performance through objective measures and human evaluation across a diverse linguistic landscape. By releasing our code and models publicly, we aim to empower communities with limited linguistic resources and foster further innovation in the field of speech technology.

Speech Loudness in Broadcasting and Streaming

The introduction and regulation of loudness in broadcasting and streaming brought clear benefits to the audience, e.g., a level of uniformity across programs and channels. Yet, speech loudness is frequently reported as being too low in certain passages, which can hinder the full understanding and enjoyment of movies and TV programs. This paper proposes expanding the set of loudness-based measures typically used in the industry. We focus on speech loudness, and we show that, when clean speech is not available, Deep Neural Networks (DNNs) can be used to isolate the speech signal and so to accurately estimate speech loudness, providing a more precise estimate compared to speech-gated loudness. Moreover, we define critical passages, i.e., passages in which speech is likely to be hard to understand. Critical passages are defined based on the local Speech Loudness Deviation (SLD) and the local Speech-to-Background Loudness Difference (SBLD), as SLD and SBLD significantly contribute to intelligibility and listening effort. In contrast to other more comprehensive measures of intelligibility and listening effort, SLD and SBLD can be straightforwardly measured, are intuitive, and, most importantly, can be easily controlled by adjusting the speech level in the mix or by enabling personalization at the user's end. Finally, examples are provided that show how the detection of critical passages can support the evaluation and control of the speech signal during and after content production.

Data-driven Joint Detection and Localization of Acoustic Reflectors

Room geometry inference algorithms rely on the localization of acoustic reflectors to identify boundary surfaces of an enclosure. Rooms with highly absorptive walls or walls at large distances from the measurement setup pose challenges for such algorithms. As it is not always possible to localize all walls, we present a data-driven method to jointly detect and localize acoustic reflectors that correspond to nearby and/or reflective walls. A multi-branch convolutional recurrent neural network is employed for this purpose. The network's input consists of a time-domain acoustic beamforming map, obtained via Radon transform from multi-channel room impulse responses. A modified loss function is proposed that forces the network to pay more attention to walls that can be estimated with a small error. Simulation results show that the proposed method can detect nearby and/or reflective walls and improve the localization performance for the detected walls.