Konstantinos Drossos

Samuel Lipping, Parthasaarathy Sudarsanam, Konstantinos Drossos et al.

Audio question answering (AQA) is a multimodal translation task where a system analyzes an audio signal and a natural language question, to generate a desirable natural language answer. In this paper, we introduce Clotho-AQA, a dataset for Audio question answering consisting of 1991 audio files each between 15 to 30 seconds in duration selected from the Clotho dataset. For each audio file, we collect six different questions and corresponding answers by crowdsourcing using Amazon Mechanical Turk. The questions and answers are produced by different annotators. Out of the six questions for each audio, two questions each are designed to have 'yes' and 'no' as answers, while the remaining two questions have other single-word answers. For each question, we collect answers from three different annotators. We also present two baseline experiments to describe the usage of our dataset for the AQA task - an LSTM-based multimodal binary classifier for 'yes' or 'no' type answers and an LSTM-based multimodal multi-class classifier for 828 single-word answers. The binary classifier achieved an accuracy of 62.7% and the multi-class classifier achieved a top-1 accuracy of 54.2% and a top-5 accuracy of 93.7%. Clotho-AQA dataset is freely available online at https://zenodo.org/record/6473207.

Shayan Gharib, Minh Tran, Diep Luong et al.

Sound event detection systems are widely used in various applications such as surveillance and environmental monitoring where data is automatically collected, processed, and sent to a cloud for sound recognition. However, this process may inadvertently reveal sensitive information about users or their surroundings, hence raising privacy concerns. In this study, we propose a novel adversarial training method for learning representations of audio recordings that effectively prevents the detection of speech activity from the latent features of the recordings. The proposed method trains a model to generate invariant latent representations of speech-containing audio recordings that cannot be distinguished from non-speech recordings by a speech classifier. The novelty of our work is in the optimization algorithm, where the speech classifier's weights are regularly replaced with the weights of classifiers trained in a supervised manner. This increases the discrimination power of the speech classifier constantly during the adversarial training, motivating the model to generate latent representations in which speech is not distinguishable, even using new speech classifiers trained outside the adversarial training loop. The proposed method is evaluated against a baseline approach with no privacy measures and a prior adversarial training method, demonstrating a significant reduction in privacy violations compared to the baseline approach. Additionally, we show that the prior adversarial method is practically ineffective for this purpose.

Diep Luong, Minh Tran, Shayan Gharib et al.

Privacy preservation has long been a concern in smart acoustic monitoring systems, where speech can be passively recorded along with a target signal in the system's operating environment. In this study, we propose the integration of two commonly used approaches in privacy preservation: source separation and adversarial representation learning. The proposed system learns the latent representation of audio recordings such that it prevents differentiating between speech and non-speech recordings. Initially, the source separation network filters out some of the privacy-sensitive data, and during the adversarial learning process, the system will learn privacy-preserving representation on the filtered signal. We demonstrate the effectiveness of our proposed method by comparing our method against systems without source separation, without adversarial learning, and without both. Overall, our results suggest that the proposed system can significantly improve speech privacy preservation compared to that of using source separation or adversarial learning solely while maintaining good performance in the acoustic monitoring task.

Diep Luong, Konstantinos Drossos, Mikko Heikkinen et al.

Audio context determines which sound components and sources are relevant and which can be perceived as irrelevant (noise) by listeners. For example, traffic noise is informative in urban surveillance but noise for a phone call at the same location. Most current audio denoising systems apply fixed target-noise definitions, often removing useful components in one context while failing to suppress irrelevant components. To address this, we introduce the concept automatic contextual audio denoising (ACAD) which defines target and noise based on the inferred context. In this work, we restrict context to be associated with an acoustic scene class. We label sound events outside the event distribution of a scene class (noise) as out-of-context (OC) and events typical for that scene as in-context (IC). We implement a deep learning method that automatically infers the context of the audio signal and removes OC components, and benchmark it against variants: without context inference, with oracle context, and with separately provided uninformative context. On paired clean/noisy data across diverse contexts, where OC components in one context may be IC in another, our proposed method outperforms other approaches across standard objective metrics, indicating that the model can infer context and context-dependent processing can enhance denoising.

Mikko Heikkinen, Archontis Politis, Konstantinos Drossos et al.

Using deep neural networks (DNNs) for encoding of microphone array (MA) signals to the Ambisonics spatial audio format can surpass certain limitations of established conventional methods, but existing DNN-based methods need to be trained separately for each MA. This paper proposes a DNN-based method for Ambisonics encoding that can generalize to arbitrary MA geometries unseen during training. The method takes as inputs the MA geometry and MA signals and uses a multi-level encoder consisting of separate paths for geometry and signal data, where geometry features inform the signal encoder at each level. The method is validated in simulated anechoic and reverberant conditions with one and two sources. The results indicate improvement over conventional encoding across the whole frequency range for dry scenes, while for reverberant scenes the improvement is frequency-dependent.

Konstantinos Drossos, Mikko Heikkinen, Paschalis Tsiaflakis

Speech denoising (SD) is an important task of many, if not all, modern signal processing chains used in devices and for everyday-life applications. While there are many published and powerful deep neural network (DNN)-based methods for SD, few are optimized for resource-constrained platforms such as mobile devices. Additionally, most DNN-based methods for SD are not focusing on full-band (FB) signals, i.e. having 48 kHz sampling rate, and/or low latency cases. In this paper we present a causal, low latency, and lightweight DNN-based method for full-band SD, leveraging both short and long temporal patterns. The method is based on a modified UNet architecture employing look-back frames, temporal spanning of convolutional kernels, and recurrent neural networks for exploiting short and long temporal patterns in the signal and estimated denoising mask. The DNN operates on a causal frame-by-frame basis taking as an input the STFT magnitude, utilizes inverted bottlenecks inspired by MobileNet, employs causal instance normalization for channel-wise normalization, and achieves a real-time factor below 0.02 when deployed on a modern mobile phone. The proposed method is evaluated using established speech denoising metrics and publicly available datasets, demonstrating its effectiveness in achieving an (SI-)SDR value that outperforms existing FB and low latency SD methods.

Diep Luong, Mikko Heikkinen, Konstantinos Drossos et al.

Speech denoising is a generally adopted and impactful task, appearing in many common and everyday-life use cases. Although there are very powerful methods published, most of those are too complex for deployment in everyday and low-resources computational environments, like hand-held devices, intelligent glasses, hearing aids, etc. Knowledge distillation (KD) is a prominent way for alleviating this complexity mismatch and is based on the transferring/distilling of knowledge from a pre-trained complex model, the teacher, to another less complex one, the student. Existing KD methods for speech denoising are based on processes that potentially hamper the KD by bounding the learning of the student to the distribution, information ordering, and feature dimensionality learned by the teacher. In this paper, we present and assess a method that tries to treat this issue, by exploiting the well-known denoising-autoencoder framework, the linear inverted bottlenecks, and the properties of the cosine similarity. We use a public dataset and conduct repeated experiments with different mismatching scenarios between the teacher and the student, reporting the mean and standard deviation of the metrics of our method and another, state-of-the-art method that is used as a baseline. Our results show that with the proposed method, the student can perform better and can also retain greater mismatching conditions compared to the teacher.

Benno Weck, Xavier Favory, Konstantinos Drossos et al.

Automated audio captioning (AAC) is the task of automatically generating textual descriptions for general audio signals. A captioning system has to identify various information from the input signal and express it with natural language. Existing works mainly focus on investigating new methods and try to improve their performance measured on existing datasets. Having attracted attention only recently, very few works on AAC study the performance of existing pre-trained audio and natural language processing resources. In this paper, we evaluate the performance of off-the-shelf models with a Transformer-based captioning approach. We utilize the freely available Clotho dataset to compare four different pre-trained machine listening models, four word embedding models, and their combinations in many different settings. Our evaluation suggests that YAMNet combined with BERT embeddings produces the best captions. Moreover, in general, fine-tuning pre-trained word embeddings can lead to better performance. Finally, we show that sequences of audio embeddings can be processed using a Transformer encoder to produce higher-quality captions.

Andreas Triantafyllopoulos, Manuel Milling, Konstantinos Drossos et al.

Underspecification and fairness in machine learning (ML) applications have recently become two prominent issues in the ML community. Acoustic scene classification (ASC) applications have so far remained unaffected by this discussion, but are now becoming increasingly used in real-world systems where fairness and reliability are critical aspects. In this work, we argue for the need of a more holistic evaluation process for ASC models through disaggregated evaluations. This entails taking into account performance differences across several factors, such as city, location, and recording device. Although these factors play a well-understood role in the performance of ASC models, most works report single evaluation metrics taking into account all different strata of a particular dataset. We argue that metrics computed on specific sub-populations of the underlying data contain valuable information about the expected real-world behaviour of proposed systems, and their reporting could improve the transparency and trustability of such systems. We demonstrate the effectiveness of the proposed evaluation process in uncovering underspecification and fairness problems exhibited by several standard ML architectures when trained on two widely-used ASC datasets. Our evaluation shows that all examined architectures exhibit large biases across all factors taken into consideration, and in particular with respect to the recording location. Additionally, different architectures exhibit different biases even though they are trained with the same experimental configurations.

Parthasaarathy Sudarsanam, Archontis Politis, Konstantinos Drossos

Joint sound event localization and detection (SELD) is an emerging audio signal processing task adding spatial dimensions to acoustic scene analysis and sound event detection. A popular approach to modeling SELD jointly is using convolutional recurrent neural network (CRNN) models, where CNNs learn high-level features from multi-channel audio input and the RNNs learn temporal relationships from these high-level features. However, RNNs have some drawbacks, such as a limited capability to model long temporal dependencies and slow training and inference times due to their sequential processing nature. Recently, a few SELD studies used multi-head self-attention (MHSA), among other innovations in their models. MHSA and the related transformer networks have shown state-of-the-art performance in various domains. While they can model long temporal dependencies, they can also be parallelized efficiently. In this paper, we study in detail the effect of MHSA on the SELD task. Specifically, we examined the effects of replacing the RNN blocks with self-attention layers. We studied the influence of stacking multiple self-attention blocks, using multiple attention heads in each self-attention block, and the effect of position embeddings and layer normalization. Evaluation on the DCASE 2021 SELD (task 3) development data set shows a significant improvement in all employed metrics compared to the baseline CRNN accompanying the task.

Jan Berg, Konstantinos Drossos

Automated audio captioning (AAC) is the task of automatically creating textual descriptions (i.e. captions) for the contents of a general audio signal. Most AAC methods are using existing datasets to optimize and/or evaluate upon. Given the limited information held by the AAC datasets, it is very likely that AAC methods learn only the information contained in the utilized datasets. In this paper we present a first approach for continuously adapting an AAC method to new information, using a continual learning method. In our scenario, a pre-optimized AAC method is used for some unseen general audio signals and can update its parameters in order to adapt to the new information, given a new reference caption. We evaluate our method using a freely available, pre-optimized AAC method and two freely available AAC datasets. We compare our proposed method with three scenarios, two of training on one of the datasets and evaluating on the other and a third of training on one dataset and fine-tuning on the other. Obtained results show that our method achieves a good balance between distilling new knowledge and not forgetting the previous one.

Einari Vaaras, Sari Ahlqvist-Björkroth, Konstantinos Drossos et al.

Researchers have recently started to study how the emotional speech heard by young infants can affect their developmental outcomes. As a part of this research, hundreds of hours of daylong recordings from preterm infants' audio environments were collected from two hospitals in Finland and Estonia in the context of so-called APPLE study. In order to analyze the emotional content of speech in such a massive dataset, an automatic speech emotion recognition (SER) system is required. However, there are no emotion labels or existing indomain SER systems to be used for this purpose. In this paper, we introduce this initially unannotated large-scale real-world audio dataset and describe the development of a functional SER system for the Finnish subset of the data. We explore the effectiveness of alternative state-of-the-art techniques to deploy a SER system to a new domain, comparing cross-corpus generalization, WGAN-based domain adaptation, and active learning in the task. As a result, we show that the best-performing models are able to achieve a classification performance of 73.4% unweighted average recall (UAR) and 73.2% UAR for a binary classification for valence and arousal, respectively. The results also show that active learning achieves the most consistent performance compared to the two alternatives.

Andres Ferraro, Xavier Favory, Konstantinos Drossos et al.

Modeling various aspects that make a music piece unique is a challenging task, requiring the combination of multiple sources of information. Deep learning is commonly used to obtain representations using various sources of information, such as the audio, interactions between users and songs, or associated genre metadata. Recently, contrastive learning has led to representations that generalize better compared to traditional supervised methods. In this paper, we present a novel approach that combines multiple types of information related to music using cross-modal contrastive learning, allowing us to learn an audio feature from heterogeneous data simultaneously. We align the latent representations obtained from playlists-track interactions, genre metadata, and the tracks' audio, by maximizing the agreement between these modality representations using a contrastive loss. We evaluate our approach in three tasks, namely, genre classification, playlist continuation and automatic tagging. We compare the performances with a baseline audio-based CNN trained to predict these modalities. We also study the importance of including multiple sources of information when training our embedding model. The results suggest that the proposed method outperforms the baseline in all the three downstream tasks and achieves comparable performance to the state-of-the-art.

Emmanuel Rovithis, Nikolaos Moustakas, Konstantinos Vogklis et al.

Citizen Science aims to engage people in research activities on important issues related to their well-being. Smart Cities aim to provide them with services that improve the quality of their life. Both concepts have seen significant growth in the last years, and can be further enhanced by combining their purposes with IoT technologies that allow for dynamic and large-scale communication and interaction. However, exciting and retaining the interest of participants is a key factor for such initiatives. In this paper we suggest that engagement in Citizen Science projects applied on Smart Cities infrastructure can be enhanced through contextual and structural game elements realized through augmented audio interactive mechanisms. Our inter-disciplinary framework is described through the paradigm of a collaborative bird call recognition game, in which users collect and submit audio data, which are then classified and used for augmenting physical space with virtual soundscape maps. We discuss the Playful Learning, Internet of Audio Things, and Bird Monitoring principles that shaped the design of our paradigm, and analyze its potential technical implementation.

Xavier Favory, Konstantinos Drossos, Tuomas Virtanen et al.

Self-supervised audio representation learning offers an attractive alternative for obtaining generic audio embeddings, capable to be employed into various downstream tasks. Published approaches that consider both audio and words/tags associated with audio do not employ text processing models that are capable to generalize to tags unknown during training. In this work we propose a method for learning audio representations using an audio autoencoder (AAE), a general word embeddings model (WEM), and a multi-head self-attention (MHA) mechanism. MHA attends on the output of the WEM, providing a contextualized representation of the tags associated with the audio, and we align the output of MHA with the output of the encoder of AAE using a contrastive loss. We jointly optimize AAE and MHA and we evaluate the audio representations (i.e. the output of the encoder of AAE) by utilizing them in three different downstream tasks, namely sound, music genre, and music instrument classification. Our results show that employing multi-head self-attention with multiple heads in the tag-based network can induce better learned audio representations.

An Tran, Konstantinos Drossos, Tuomas Virtanen

Automated audio captioning (AAC) is a novel task, where a method takes as an input an audio sample and outputs a textual description (i.e. a caption) of its contents. Most AAC methods are adapted from from image captioning of machine translation fields. In this work we present a novel AAC novel method, explicitly focused on the exploitation of the temporal and time-frequency patterns in audio. We employ three learnable processes for audio encoding, two for extracting the local and temporal information, and one to merge the output of the previous two processes. To generate the caption, we employ the widely used Transformer decoder. We assess our method utilizing the freely available splits of Clotho dataset. Our results increase previously reported highest SPIDEr to 17.3, from 16.2.

Konstantinos Drossos, Stylianos I. Mimilakis, Tuomas Virtanen

Sound event detection (SED) is the task of identifying sound events along with their onset and offset times. A recent, convolutional neural networks based SED method, proposed the usage of depthwise separable (DWS) and time-dilated convolutions. DWS and time-dilated convolutions yielded state-of-the-art results for SED, with considerable small amount of parameters. In this work we propose the expansion of the time-dilated convolutions, by conditioning them with jointly learned embeddings of the SED predictions by the SED classifier. We present a novel algorithm for the conditioning of the time-dilated convolutions which functions similarly to language modelling, and enhances the performance of the these convolutions. We employ the freely available TUT-SED Synthetic dataset, and we assess the performance of our method using the average per-frame $\text{F}_{1}$ score and average per-frame error rate, over the 10 experiments. We achieve an increase of 2\% (from 0.63 to 0.65) at the average $\text{F}_{1}$ score (the higher the better) and a decrease of 3\% (from 0.50 to 0.47) at the error rate (the lower the better).

Emre Çakır, Konstantinos Drossos, Tuomas Virtanen

Audio captioning is a multi-modal task, focusing on using natural language for describing the contents of general audio. Most audio captioning methods are based on deep neural networks, employing an encoder-decoder scheme and a dataset with audio clips and corresponding natural language descriptions (i.e. captions). A significant challenge for audio captioning is the distribution of words in the captions: some words are very frequent but acoustically non-informative, i.e. the function words (e.g. "a", "the"), and other words are infrequent but informative, i.e. the content words (e.g. adjectives, nouns). In this paper we propose two methods to mitigate this class imbalance problem. First, in an autoencoder setting for audio captioning, we weigh each word's contribution to the training loss inversely proportional to its number of occurrences in the whole dataset. Secondly, in addition to multi-class, word-level audio captioning task, we define a multi-label side task based on clip-level content word detection by training a separate decoder. We use the loss from the second task to regularize the jointly trained encoder for the audio captioning task. We evaluate our method using Clotho, a recently published, wide-scale audio captioning dataset, and our results show an increase of 37\% relative improvement with SPIDEr metric over the baseline method.

Stylianos Ioannis Mimilakis, Konstantinos Drossos, Gerald Schuller

In this work we present a method for unsupervised learning of audio representations, focused on the task of singing voice separation. We build upon a previously proposed method for learning representations of time-domain music signals with a re-parameterized denoising autoencoder, extending it by using the family of Sinkhorn distances with entropic regularization. We evaluate our method on the freely available MUSDB18 dataset of professionally produced music recordings, and our results show that Sinkhorn distances with small strength of entropic regularization are marginally improving the performance of informed singing voice separation. By increasing the strength of the entropic regularization, the learned representations of the mixture signal consists of almost perfectly additive and distinctly structured sources.

Pyry Pyykkönen, Styliannos I. Mimilakis, Konstantinos Drossos et al.

Recent approaches for music source separation are almost exclusively based on deep neural networks, mostly employing recurrent neural networks (RNNs). Although RNNs are in many cases superior than other types of deep neural networks for sequence processing, they are known to have specific difficulties in training and parallelization, especially for the typically long sequences encountered in music source separation. In this paper we present a use-case of replacing RNNs with depth-wise separable (DWS) convolutions, which are a lightweight and faster variant of the typical convolutions. We focus on singing voice separation, employing an RNN architecture, and we replace the RNNs with DWS convolutions (DWS-CNNs). We conduct an ablation study and examine the effect of the number of channels and layers of DWS-CNNs on the source separation performance, by utilizing the standard metrics of signal-to-artifacts, signal-to-interference, and signal-to-distortion ratio. Our results show that by replacing RNNs with DWS-CNNs yields an improvement of 1.20, 0.06, 0.37 dB, respectively, while using only 20.57% of the amount of parameters of the RNN architecture.

Khoa Nguyen, Konstantinos Drossos, Tuomas Virtanen

Audio captioning is the task of automatically creating a textual description for the contents of a general audio signal. Typical audio captioning methods rely on deep neural networks (DNNs), where the target of the DNN is to map the input audio sequence to an output sequence of words, i.e. the caption. Though, the length of the textual description is considerably less than the length of the audio signal, for example 10 words versus some thousands of audio feature vectors. This clearly indicates that an output word corresponds to multiple input feature vectors. In this work we present an approach that focuses on explicitly taking advantage of this difference of lengths between sequences, by applying a temporal sub-sampling to the audio input sequence. We employ a sequence-to-sequence method, which uses a fixed-length vector as an output from the encoder, and we apply temporal sub-sampling between the RNNs of the encoder. We evaluate the benefit of our approach by employing the freely available dataset Clotho and we evaluate the impact of different factors of temporal sub-sampling. Our results show an improvement to all considered metrics.

Xavier Favory, Konstantinos Drossos, Tuomas Virtanen et al.

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.

Stylianos I. Mimilakis, Konstantinos Drossos, Gerald Schuller

In this work, we present a method for learning interpretable music signal representations directly from waveform signals. Our method can be trained using unsupervised objectives and relies on the denoising auto-encoder model that uses a simple sinusoidal model as decoding functions to reconstruct the singing voice. To demonstrate the benefits of our method, we employ the obtained representations to the task of informed singing voice separation via binary masking, and measure the obtained separation quality by means of scale-invariant signal to distortion ratio. Our findings suggest that our method is capable of learning meaningful representations for singing voice separation, while preserving conveniences of the the short-time Fourier transform like non-negativity, smoothness, and reconstruction subject to time-frequency masking, that are desired in audio and music source separation.

Antonio J. Muñoz-Montoro, Julio J. Carabias-Orti, Archontis Politis et al.

This work addresses the problem of multichannel source separation combining two powerful approaches, multichannel spectral factorization with recent monophonic deep-learning (DL) based spectrum inference. Individual source spectra at different channels are estimated with a Masker-Denoiser Twin Network (MaD TwinNet), able to model long-term temporal patterns of a musical piece. The monophonic source spectrograms are used within a spatial covariance mixing model based on Complex Non-Negative Matrix Factorization (CNMF) that predicts the spatial characteristics of each source. The proposed framework is evaluated on the task of singing voice separation with a large multichannel dataset. Experimental results show that our joint DL+CNMF method outperforms both the individual monophonic DL-based separation and the multichannel CNMF baseline methods.

Konstantinos Drossos, Stylianos I. Mimilakis, Shayan Gharib et al.

State-of-the-art sound event detection (SED) methods usually employ a series of convolutional neural networks (CNNs) to extract useful features from the input audio signal, and then recurrent neural networks (RNNs) to model longer temporal context in the extracted features. The number of the channels of the CNNs and size of the weight matrices of the RNNs have a direct effect on the total amount of parameters of the SED method, which is to a couple of millions. Additionally, the usually long sequences that are used as an input to an SED method along with the employment of an RNN, introduce implications like increased training time, difficulty at gradient flow, and impeding the parallelization of the SED method. To tackle all these problems, we propose the replacement of the CNNs with depthwise separable convolutions and the replacement of the RNNs with dilated convolutions. We compare the proposed method to a baseline convolutional neural network on a SED task, and achieve a reduction of the amount of parameters by 85% and average training time per epoch by 78%, and an increase the average frame-wise F1 score and reduction of the average error rate by 4.6% and 3.8%, respectively.

Shayan Gharib, Konstantinos Drossos, Eemi Fagerlund et al.

The performance of sound event detection methods can significantly degrade when they are used in unseen conditions (e.g. recording devices, ambient noise). Domain adaptation is a promising way to tackle this problem. In this paper, we present VOICe, the first dataset for the development and evaluation of domain adaptation methods for sound event detection. VOICe consists of mixtures with three different sound events ("baby crying", "glass breaking", and "gunshot"), which are over-imposed over three different categories of acoustic scenes: vehicle, outdoors, and indoors. Moreover, the mixtures are also offered without any background noise. VOICe is freely available online (https://doi.org/10.5281/zenodo.3514950). In addition, using an adversarial-based training method, we evaluate the performance of a domain adaptation method on VOICe.

Niccoló Nicodemo, Gaurav Naithani, Konstantinos Drossos et al.

Effective employment of deep neural networks (DNNs) in mobile devices and embedded systems is hampered by requirements for memory and computational power. This paper presents a non-uniform quantization approach which allows for dynamic quantization of DNN parameters for different layers and within the same layer. A virtual bit shift (VBS) scheme is also proposed to improve the accuracy of the proposed scheme. Our method reduces the memory requirements, preserving the performance of the network. The performance of our method is validated in a speech enhancement application, where a fully connected DNN is used to predict the clean speech spectrum from the input noisy speech spectrum. A DNN is optimized and its memory footprint and performance are evaluated using the short-time objective intelligibility, STOI, metric. The application of the low-bit quantization allows a 50% reduction of the DNN memory footprint while the STOI performance drops only by 2.7%.

Konstantinos Drossos, Samuel Lipping, Tuomas Virtanen

Audio captioning is the novel task of general audio content description using free text. It is an intermodal translation task (not speech-to-text), where a system accepts as an input an audio signal and outputs the textual description (i.e. the caption) of that signal. In this paper we present Clotho, a dataset for audio captioning consisting of 4981 audio samples of 15 to 30 seconds duration and 24 905 captions of eight to 20 words length, and a baseline method to provide initial results. Clotho is built with focus on audio content and caption diversity, and the splits of the data are not hampering the training or evaluation of methods. All sounds are from the Freesound platform, and captions are crowdsourced using Amazon Mechanical Turk and annotators from English speaking countries. Unique words, named entities, and speech transcription are removed with post-processing. Clotho is freely available online (https://zenodo.org/record/3490684).

Samuel Lipping, Konstantinos Drossos, Tuomas Virtanen

Audio captioning is a novel field of multi-modal translation and it is the task of creating a textual description of the content of an audio signal (e.g. "people talking in a big room"). The creation of a dataset for this task requires a considerable amount of work, rendering the crowdsourcing a very attractive option. In this paper we present a three steps based framework for crowdsourcing an audio captioning dataset, based on concepts and practises followed for the creation of widely used image captioning and machine translations datasets. During the first step initial captions are gathered. A grammatically corrected and/or rephrased version of each initial caption is obtained in second step. Finally, the initial and edited captions are rated, keeping the top ones for the produced dataset. We objectively evaluate the impact of our framework during the process of creating an audio captioning dataset, in terms of diversity and amount of typographical errors in the obtained captions. The obtained results show that the resulting dataset has less typographical errors than the initial captions, and on average each sound in the produced dataset has captions with a Jaccard similarity of 0.24, roughly equivalent to two ten-word captions having in common four words with the same root, indicating that the captions are dissimilar while they still contain some of the same information.

Konstantinos Drossos, Shayan Gharib, Paul Magron et al.

A sound event detection (SED) method typically takes as an input a sequence of audio frames and predicts the activities of sound events in each frame. In real-life recordings, the sound events exhibit some temporal structure: for instance, a "car horn" will likely be followed by a "car passing by". While this temporal structure is widely exploited in sequence prediction tasks (e.g., in machine translation), where language models (LM) are exploited, it is not satisfactorily modeled in SED. In this work we propose a method which allows a recurrent neural network (RNN) to learn an LM for the SED task. The method conditions the input of the RNN with the activities of classes at the previous time step. We evaluate our method using F1 score and error rate (ER) over three different and publicly available datasets; the TUT-SED Synthetic 2016 and the TUT Sound Events 2016 and 2017 datasets. The obtained results show an increase of 9% and 2% at the F1 (higher is better) and a decrease of 7% and 2% at ER (lower is better) for the TUT Sound Events 2016 and 2017 datasets, respectively, when using our method. On the contrary, with our method there is a decrease of 4% at F1 score and an increase of 7% at ER for the TUT-SED Synthetic 2016 dataset.

Konstantinos Drossos, Paul Magron, Tuomas Virtanen

A challenging problem in deep learning-based machine listening field is the degradation of the performance when using data from unseen conditions. In this paper we focus on the acoustic scene classification (ASC) task and propose an adversarial deep learning method to allow adapting an acoustic scene classification system to deal with a new acoustic channel resulting from data captured with a different recording device. We build upon the theoretical model of HΔH-distance and previous adversarial discriminative deep learning method for ASC unsupervised domain adaptation, and we present an adversarial training based method using the Wasserstein distance. We improve the state-of-the-art mean accuracy on the data from the unseen conditions from 32% to 45%, using the TUT Acoustic Scenes dataset.

Stylianos Ioannis Mimilakis, Konstantinos Drossos, Estefanía Cano et al.

The goal of this work is to investigate what singing voice separation approaches based on neural networks learn from the data. We examine the mapping functions of neural networks based on the denoising autoencoder (DAE) model that are conditioned on the mixture magnitude spectra. To approximate the mapping functions, we propose an algorithm inspired by the knowledge distillation, denoted the neural couplings algorithm (NCA). The NCA yields a matrix that expresses the mapping of the mixture to the target source magnitude information. Using the NCA, we examine the mapping functions of three fundamental DAE-based models in music source separation; one with single-layer encoder and decoder, one with multi-layer encoder and single-layer decoder, and one using skip-filtering connections (SF) with a single-layer encoding and decoding. We first train these models with realistic data to estimate the singing voice magnitude spectra from the corresponding mixture. We then use the optimized models and test spectral data as input to the NCA. Our experimental findings show that approaches based on the DAE model learn scalar filtering operators, exhibiting a predominant diagonal structure in their corresponding mapping functions, limiting the exploitation of inter-frequency structure of music data. In contrast, skip-filtering connections are shown to assist the DAE model in learning filtering operators that exploit richer inter-frequency structures.

Shayan Gharib, Konstantinos Drossos, Emre Çakir et al.

A general problem in acoustic scene classification task is the mismatched conditions between training and testing data, which significantly reduces the performance of the developed methods on classification accuracy. As a countermeasure, we present the first method of unsupervised adversarial domain adaptation for acoustic scene classification. We employ a model pre-trained on data from one set of conditions and by using data from other set of conditions, we adapt the model in order that its output cannot be used for classifying the set of conditions that input data belong to. We use a freely available dataset from the DCASE 2018 challenge Task 1, subtask B, that contains data from mismatched recording devices. We consider the scenario where the annotations are available for the data recorded from one device, but not for the rest. Our results show that with our model agnostic method we can achieve $\sim 10\%$ increase at the accuracy on an unseen and unlabeled dataset, while keeping almost the same performance on the labeled dataset.

Konstantinos Drossos, Paul Magron, Stylianos Ioannis Mimilakis et al.

Harmonic/percussive source separation (HPSS) consists in separating the pitched instruments from the percussive parts in a music mixture. In this paper, we propose to apply the recently introduced Masker-Denoiser with twin networks (MaD TwinNet) system to this task. MaD TwinNet is a deep learning architecture that has reached state-of-the-art results in monaural singing voice separation. Herein, we propose to apply it to HPSS by using it to estimate the magnitude spectrogram of the percussive source. Then, we retrieve the complex-valued short-time Fourier transform of the sources by means of a phase recovery algorithm, which minimizes the reconstruction error and enforces the phase of the harmonic part to follow a sinusoidal phase model. Experiments conducted on realistic music mixtures show that this novel separation system outperforms the previous state-of-the art kernel additive model approach.

Konstantinos Drossos, Stylianos Ioannis Mimilakis, Andreas Floros et al.

Close miking represents a widely employed practice of placing a microphone very near to the sound source in order to capture more direct sound and minimize any pickup of ambient sound, including other, concurrently active sources. It is used by the audio engineering community for decades for audio recording, based on a number of empirical rules that were evolved during the recording practice itself. But can this empirical knowledge and close miking practice be systematically verified? In this work we aim to address this question based on an analytic methodology that employs techniques and metrics originating from the sound source separation evaluation field. In particular, we apply a quantitative analysis of the source separation capabilities of the close miking technique. The analysis is applied on a recording dataset obtained at multiple positions of a typical musical hall, multiple distances between the microphone and the sound source multiple microphone types and multiple level differences between the sound source and the ambient acoustic component. For all the above cases we compute the Source to Interference Ratio (SIR) metric. The results obtained clearly demonstrate an optimum close-miking performance that matches the current empirical knowledge of professional audio recording.

Konstantinos Drossos, Stylianos Ioannis Mimilakis, Dmitriy Serdyuk et al.

Monaural singing voice separation task focuses on the prediction of the singing voice from a single channel music mixture signal. Current state of the art (SOTA) results in monaural singing voice separation are obtained with deep learning based methods. In this work we present a novel deep learning based method that learns long-term temporal patterns and structures of a musical piece. We build upon the recently proposed Masker-Denoiser (MaD) architecture and we enhance it with the Twin Networks, a technique to regularize a recurrent generative network using a backward running copy of the network. We evaluate our method using the Demixing Secret Dataset and we obtain an increment to signal-to-distortion ratio (SDR) of 0.37 dB and to signal-to-interference ratio (SIR) of 0.23 dB, compared to previous SOTA results.

Stylianos Ioannis Mimilakis, Konstantinos Drossos, João F. Santos et al.

Singing voice separation based on deep learning relies on the usage of time-frequency masking. In many cases the masking process is not a learnable function or is not encapsulated into the deep learning optimization. Consequently, most of the existing methods rely on a post processing step using the generalized Wiener filtering. This work proposes a method that learns and optimizes (during training) a source-dependent mask and does not need the aforementioned post processing step. We introduce a recurrent inference algorithm, a sparse transformation step to improve the mask generation process, and a learned denoising filter. Obtained results show an increase of 0.49 dB for the signal to distortion ratio and 0.30 dB for the signal to interference ratio, compared to previous state-of-the-art approaches for monaural singing voice separation.

Stylianos Ioannis Mimilakis, Konstantinos Drossos, Tuomas Virtanen et al.

The objective of deep learning methods based on encoder-decoder architectures for music source separation is to approximate either ideal time-frequency masks or spectral representations of the target music source(s). The spectral representations are then used to derive time-frequency masks. In this work we introduce a method to directly learn time-frequency masks from an observed mixture magnitude spectrum. We employ recurrent neural networks and train them using prior knowledge only for the magnitude spectrum of the target source. To assess the performance of the proposed method, we focus on the task of singing voice separation. The results from an objective evaluation show that our proposed method provides comparable results to deep learning based methods which operate over complicated signal representations. Compared to previous methods that approximate time-frequency masks, our method has increased performance of signal to distortion ratio by an average of 3.8 dB.

Konstantinos Drossos, Sharath Adavanne, Tuomas Virtanen

We present the first approach to automated audio captioning. We employ an encoder-decoder scheme with an alignment model in between. The input to the encoder is a sequence of log mel-band energies calculated from an audio file, while the output is a sequence of words, i.e. a caption. The encoder is a multi-layered, bi-directional gated recurrent unit (GRU) and the decoder a multi-layered GRU with a classification layer connected to the last GRU of the decoder. The classification layer and the alignment model are fully connected layers with shared weights between timesteps. The proposed method is evaluated using data drawn from a commercial sound effects library, ProSound Effects. The resulting captions were rated through metrics utilized in machine translation and image captioning fields. Results from metrics show that the proposed method can predict words appearing in the original caption, but not always correctly ordered.

Miroslav Malik, Sharath Adavanne, Konstantinos Drossos et al.

This paper studies the emotion recognition from musical tracks in the 2-dimensional valence-arousal (V-A) emotional space. We propose a method based on convolutional (CNN) and recurrent neural networks (RNN), having significantly fewer parameters compared with the state-of-the-art method for the same task. We utilize one CNN layer followed by two branches of RNNs trained separately for arousal and valence. The method was evaluated using the 'MediaEval2015 emotion in music' dataset. We achieved an RMSE of 0.202 for arousal and 0.268 for valence, which is the best result reported on this dataset.

Sharath Adavanne, Konstantinos Drossos, Emre Çakır et al.

This paper studies the detection of bird calls in audio segments using stacked convolutional and recurrent neural networks. Data augmentation by blocks mixing and domain adaptation using a novel method of test mixing are proposed and evaluated in regard to making the method robust to unseen data. The contributions of two kinds of acoustic features (dominant frequency and log mel-band energy) and their combinations are studied in the context of bird audio detection. Our best achieved AUC measure on five cross-validations of the development data is 95.5% and 88.1% on the unseen evaluation data.

EmreÇakır, Sharath Adavanne, Giambattista Parascandolo et al.

Bird sounds possess distinctive spectral structure which may exhibit small shifts in spectrum depending on the bird species and environmental conditions. In this paper, we propose using convolutional recurrent neural networks on the task of automated bird audio detection in real-life environments. In the proposed method, convolutional layers extract high dimensional, local frequency shift invariant features, while recurrent layers capture longer term dependencies between the features extracted from short time frames. This method achieves 88.5% Area Under ROC Curve (AUC) score on the unseen evaluation data and obtains the second place in the Bird Audio Detection challenge.