Yannis Stylianou

Zakhar Shumaylov, Vasileios Tsiaras, Yannis Stylianou

The ever-increasing parameter counts of deep learning models necessitate effective compression techniques for deployment on resource-constrained devices. This paper explores the application of information geometry, the study of density-induced metrics on parameter spaces, to analyze existing methods within the space of model compression, primarily focusing on operator factorization. Adopting this perspective highlights the core challenge: defining an optimal low-compute submanifold (or subset) and projecting onto it. We argue that many successful model compression approaches can be understood as implicitly approximating information divergences for this projection. We highlight that when compressing a pre-trained model, using information divergences is paramount for achieving improved zero-shot accuracy, yet this may no longer be the case when the model is fine-tuned. In such scenarios, trainability of bottlenecked models turns out to be far more important for achieving high compression ratios with minimal performance degradation, necessitating adoption of iterative methods. In this context, we prove convergence of iterative singular value thresholding for training neural networks subject to a soft rank constraint. To further illustrate the utility of this perspective, we showcase how simple modifications to existing methods through softer rank reduction result in improved performance under fixed compression rates.

Javier Latorre, Charlotte Bailleul, Tuuli Morrill et al.

In this work, we explore multiple architectures and training procedures for developing a multi-speaker and multi-lingual neural TTS system with the goals of a) improving the quality when the available data in the target language is limited and b) enabling cross-lingual synthesis. We report results from a large experiment using 30 speakers in 8 different languages across 15 different locales. The system is trained on the same amount of data per speaker. Compared to a single-speaker model, when the suggested system is fine tuned to a speaker, it produces significantly better quality in most of the cases while it only uses less than $40\%$ of the speaker's data used to build the single-speaker model. In cross-lingual synthesis, on average, the generated quality is within $80\%$ of native single-speaker models, in terms of Mean Opinion Score.

Muhammed PV Shifas, Anna Sfakianaki, Theognosia Chimona et al.

In this work we evaluate a neural based speech intelligibility booster based on spectral shaping and dynamic range compression (SSDRC), referred to as WaveNet-based SSDRC (wSSDRC), using a recently designed Greek Harvard-style corpus. The corpus has been developed according to the format of the Harvard/IEEE sentences and offers the opportunity to apply neural speech enhancement models and examine their performance gain for Greek listeners. wSSDRC has been successfully tested for English material and speakers in the past. In this paper we revisit wSSDRC to perform a full scale evaluation of the model with Greek listeners under the condition of equal energy before and after modification. Both normal hearing (NH) and hearing impaired (HI) listeners evaluated the model under speech shaped noise (SSN) at listener-specific SNRs matching their Speech Reception Threshold (SRT) - a point at which 50 % of unmodified speech is intelligible. The analysis statistics show that the wSSDRC model has produced a median intelligibility boost of 39% for NH and 38% for HI, relative to the plain unprocessed speech.

Dipjyoti Paul, Muhammed PV Shifas, Yannis Pantazis et al.

The increased adoption of digital assistants makes text-to-speech (TTS) synthesis systems an indispensable feature of modern mobile devices. It is hence desirable to build a system capable of generating highly intelligible speech in the presence of noise. Past studies have investigated style conversion in TTS synthesis, yet degraded synthesized quality often leads to worse intelligibility. To overcome such limitations, we proposed a novel transfer learning approach using Tacotron and WaveRNN based TTS synthesis. The proposed speech system exploits two modification strategies: (a) Lombard speaking style data and (b) Spectral Shaping and Dynamic Range Compression (SSDRC) which has been shown to provide high intelligibility gains by redistributing the signal energy on the time-frequency domain. We refer to this extension as Lombard-SSDRC TTS system. Intelligibility enhancement as quantified by the Intelligibility in Bits (SIIB-Gauss) measure shows that the proposed Lombard-SSDRC TTS system shows significant relative improvement between 110% and 130% in speech-shaped noise (SSN), and 47% to 140% in competing-speaker noise (CSN) against the state-of-the-art TTS approach. Additional subjective evaluation shows that Lombard-SSDRC TTS successfully increases the speech intelligibility with relative improvement of 455% for SSN and 104% for CSN in median keyword correction rate compared to the baseline TTS method.

Dipjyoti Paul, Yannis Pantazis, Yannis Stylianou

Recent advancements in deep learning led to human-level performance in single-speaker speech synthesis. However, there are still limitations in terms of speech quality when generalizing those systems into multiple-speaker models especially for unseen speakers and unseen recording qualities. For instance, conventional neural vocoders are adjusted to the training speaker and have poor generalization capabilities to unseen speakers. In this work, we propose a variant of WaveRNN, referred to as speaker conditional WaveRNN (SC-WaveRNN). We target towards the development of an efficient universal vocoder even for unseen speakers and recording conditions. In contrast to standard WaveRNN, SC-WaveRNN exploits additional information given in the form of speaker embeddings. Using publicly-available data for training, SC-WaveRNN achieves significantly better performance over baseline WaveRNN on both subjective and objective metrics. In MOS, SC-WaveRNN achieves an improvement of about 23% for seen speaker and seen recording condition and up to 95% for unseen speaker and unseen condition. Finally, we extend our work by implementing a multi-speaker text-to-speech (TTS) synthesis similar to zero-shot speaker adaptation. In terms of performance, our system has been preferred over the baseline TTS system by 60% over 15.5% and by 60.9% over 32.6%, for seen and unseen speakers, respectively.

Ahmed Hussen Abdelaziz, Anushree Prasanna Kumar, Chloe Seivwright et al.

Audiovisual speech synthesis is the problem of synthesizing a talking face while maximizing the coherency of the acoustic and visual speech. In this paper, we propose and compare two audiovisual speech synthesis systems for 3D face models. The first system is the AVTacotron2, which is an end-to-end text-to-audiovisual speech synthesizer based on the Tacotron2 architecture. AVTacotron2 converts a sequence of phonemes representing the sentence to synthesize into a sequence of acoustic features and the corresponding controllers of a face model. The output acoustic features are used to condition a WaveRNN to reconstruct the speech waveform, and the output facial controllers are used to generate the corresponding video of the talking face. The second audiovisual speech synthesis system is modular, where acoustic speech is synthesized from text using the traditional Tacotron2. The reconstructed acoustic speech signal is then used to drive the facial controls of the face model using an independently trained audio-to-facial-animation neural network. We further condition both the end-to-end and modular approaches on emotion embeddings that encode the required prosody to generate emotional audiovisual speech. We analyze the performance of the two systems and compare them to the ground truth videos using subjective evaluation tests. The end-to-end and modular systems are able to synthesize close to human-like audiovisual speech with mean opinion scores (MOS) of 4.1 and 3.9, respectively, compared to a MOS of 4.1 for the ground truth generated from professionally recorded videos. While the end-to-end system gives a better overall quality, the modular approach is more flexible and the quality of acoustic speech and visual speech synthesis is almost independent of each other.

Yannis Pantazis, Dipjyoti Paul, Michail Fasoulakis et al.

In this paper, we propose a novel loss function for training Generative Adversarial Networks (GANs) aiming towards deeper theoretical understanding as well as improved stability and performance for the underlying optimization problem. The new loss function is based on cumulant generating functions giving rise to \emph{Cumulant GAN}. Relying on a recently-derived variational formula, we show that the corresponding optimization problem is equivalent to R{é}nyi divergence minimization, thus offering a (partially) unified perspective of GAN losses: the R{é}nyi family encompasses Kullback-Leibler divergence (KLD), reverse KLD, Hellinger distance and $χ^2$-divergence. Wasserstein GAN is also a member of cumulant GAN. In terms of stability, we rigorously prove the linear convergence of cumulant GAN to the Nash equilibrium for a linear discriminator, Gaussian distributions and the standard gradient descent ascent algorithm. Finally, we experimentally demonstrate that image generation is more robust relative to Wasserstein GAN and it is substantially improved in terms of both inception score and Fréchet inception distance when both weaker and stronger discriminators are considered.

Muhammed PV Shifas, Santelli Claudio, Vassilis Tsiaras et al.

Convolutional neural network (CNN) modules are widely being used to build high-end speech enhancement neural models. However, the feature extraction power of vanilla CNN modules has been limited by the dimensionality constraint of the convolution kernels that are integrated - thereby, they have limitations to adequately model the noise context information at the feature extraction stage. To this end, adding recurrency factor into the feature extracting CNN layers, we introduce a robust context-aware feature extraction strategy for single-channel speech enhancement. As shown, adding recurrency results in capturing the local statistics of noise attributes at the extracted features level and thus, the suggested model is effective in differentiating speech cues even at very noisy conditions. When evaluated against enhancement models using vanilla CNN modules, in unseen noise conditions, the suggested model with recurrency in the feature extraction layers has produced a segmental SNR (SSNR) gain of up to 1.5 dB, an improvement of 0.4 in subjective quality in the Mean Opinion Score scale, while the parameters to be optimized are reduced by 25%.

Muhammed PV Shifas, Nagaraj Adiga, Vassilis Tsiaras et al.

In this paper, we suggest a new parallel, non-causal and shallow waveform domain architecture for speech enhancement based on FFTNet, a neural network for generating high quality audio waveform. In contrast to other waveform based approaches like WaveNet, FFTNet uses an initial wide dilation pattern. Such an architecture better represents the long term correlated structure of speech in the time domain, where noise is usually highly non-correlated, and therefore it is suitable for waveform domain based speech enhancement. To further strengthen this feature of FFTNet, we suggest a non-causal FFTNet architecture, where the present sample in each layer is estimated from the past and future samples of the previous layer. By suggesting a shallow network and applying non-causality within certain limits, the suggested FFTNet for speech enhancement (SE-FFTNet) uses much fewer parameters compared to other neural network based approaches for speech enhancement like WaveNet and SEGAN. Specifically, the suggested network has considerably reduced model parameters: 32% fewer compared to WaveNet and 87% fewer compared to SEGAN. Finally, based on subjective and objective metrics, SE-FFTNet outperforms WaveNet in terms of enhanced signal quality, while it provides equally good performance as SEGAN. A Tensorflow implementation of the architecture is provided at 1 .

Thomas Drugman, Yannis Stylianou

Maximum Voiced Frequency (MVF) is used in various speech models as the spectral boundary separating periodic and aperiodic components during the production of voiced sounds. Recent studies have shown that its proper estimation and modeling enhance the quality of statistical parametric speech synthesizers. Contrastingly, these same methods of MVF estimation have been reported to degrade the performance of singing voice synthesizers. This paper proposes a new approach for MVF estimation which exploits both amplitude and phase spectra. It is shown that phase conveys relevant information about the harmonicity of the voice signal, and that it can be jointly used with features derived from the amplitude spectrum. This information is further integrated into a maximum likelihood criterion which provides a decision about the MVF estimate. The proposed technique is compared to two state-of-the-art methods, and shows a superior performance in both objective and subjective evaluations. Perceptual tests indicate a drastic improvement in high-pitched voices.

Thomas Drugman, Yannis Stylianou, Yusuke Kida et al.

Voice Activity Detection (VAD) refers to the problem of distinguishing speech segments from background noise. Numerous approaches have been proposed for this purpose. Some are based on features derived from the power spectral density, others exploit the periodicity of the signal. The goal of this paper is to investigate the joint use of source and filter-based features. Interestingly, a mutual information-based assessment shows superior discrimination power for the source-related features, especially the proposed ones. The features are further the input of an artificial neural network-based classifier trained on a multi-condition database. Two strategies are proposed to merge source and filter information: feature and decision fusion. Our experiments indicate an absolute reduction of 3% of the equal error rate when using decision fusion. The final proposed system is compared to four state-of-the-art methods on 150 minutes of data recorded in real environments. Thanks to the robustness of its source-related features, its multi-condition training and its efficient information fusion, the proposed system yields over the best state-of-the-art VAD a substantial increase of accuracy across all conditions (24% absolute on average).

Yannis Pantazis, Dipjyoti Paul, Michail Fasoulakis et al.

The impressive success of Generative Adversarial Networks (GANs) is often overshadowed by the difficulties in their training. Despite the continuous efforts and improvements, there are still open issues regarding their convergence properties. In this paper, we propose a simple training variation where suitable weights are defined and assist the training of the Generator. We provide theoretical arguments why the proposed algorithm is better than the baseline training in the sense of speeding up the training process and of creating a stronger Generator. Performance results showed that the new algorithm is more accurate in both synthetic and image datasets resulting in improvements ranging between 5% and 50%.

Dan Stowell, Yannis Stylianou, Mike Wood et al.

Assessing the presence and abundance of birds is important for monitoring specific species as well as overall ecosystem health. Many birds are most readily detected by their sounds, and thus passive acoustic monitoring is highly appropriate. Yet acoustic monitoring is often held back by practical limitations such as the need for manual configuration, reliance on example sound libraries, low accuracy, low robustness, and limited ability to generalise to novel acoustic conditions. Here we report outcomes from a collaborative data challenge showing that with modern machine learning including deep learning, general-purpose acoustic bird detection can achieve very high retrieval rates in remote monitoring data --- with no manual recalibration, and no pre-training of the detector for the target species or the acoustic conditions in the target environment. Multiple methods were able to attain performance of around 88% AUC (area under the ROC curve), much higher performance than previous general-purpose methods. We present new acoustic monitoring datasets, summarise the machine learning techniques proposed by challenge teams, conduct detailed performance evaluation, and discuss how such approaches to detection can be integrated into remote monitoring projects.

Alexandros Papangelis, Panagiotis Papadakos, Margarita Kotti et al.

In this work we discuss the related challenges and describe an approach towards the fusion of state-of-the-art technologies from the Spoken Dialogue Systems (SDS) and the Semantic Web and Information Retrieval domains. We envision a dialogue system named LD-SDS that will support advanced, expressive, and engaging user requests, over multiple, complex, rich, and open-domain data sources that will leverage the wealth of the available Linked Data. Specifically, we focus on: a) improving the identification, disambiguation and linking of entities occurring in data sources and user input; b) offering advanced query services for exploiting the semantics of the data, with reasoning and exploratory capabilities; and c) expanding the typical information seeking dialogue model (slot filling) to better reflect real-world conversational search scenarios.

Dan Stowell, Mike Wood, Yannis Stylianou et al.

Many biological monitoring projects rely on acoustic detection of birds. Despite increasingly large datasets, this detection is often manual or semi-automatic, requiring manual tuning/postprocessing. We review the state of the art in automatic bird sound detection, and identify a widespread need for tuning-free and species-agnostic approaches. We introduce new datasets and an IEEE research challenge to address this need, to make possible the development of fully automatic algorithms for bird sound detection.