Yin Cao

Xubo Liu, Zhongkai Zhu, Haohe Liu et al.

Despite breakthroughs in audio generation models, their capabilities are often confined to domain-specific conditions such as speech transcriptions and audio captions. However, real-world audio creation aims to generate harmonious audio containing various elements such as speech, music, and sound effects with controllable conditions, which is challenging to address using existing audio generation systems. We present WavJourney, a novel framework that leverages Large Language Models (LLMs) to connect various audio models for audio creation. WavJourney allows users to create storytelling audio content with diverse audio elements simply from textual descriptions. Specifically, given a text instruction, WavJourney first prompts LLMs to generate an audio script that serves as a structured semantic representation of audio elements. The audio script is then converted into a computer program, where each line of the program calls a task-specific audio generation model or computational operation function. The computer program is then executed to obtain a compositional and interpretable solution for audio creation. Experimental results suggest that WavJourney is capable of synthesizing realistic audio aligned with textually-described semantic, spatial and temporal conditions, achieving state-of-the-art results on text-to-audio generation benchmarks. Additionally, we introduce a new multi-genre story benchmark. Subjective evaluations demonstrate the potential of WavJourney in crafting engaging storytelling audio content from text. We further demonstrate that WavJourney can facilitate human-machine co-creation in multi-round dialogues. To foster future research, the code and synthesized audio are available at: https://audio-agi.github.io/WavJourney_demopage/.

Liuyu Wu, Rui Feng, Jie Li et al.

The automated evaluation of cognitive status utilizing multimedia technologies presents a promising frontier in early dementia diagnosis. However, the development of robust machine learning models for cognitive impairment detection is frequently hindered by the scarcity of large-scale, strictly synchronized, and clinically validated multimodal datasets. To bridge this critical gap, we introduce the CogPic database, a comprehensive multimodal benchmark meticulously designed for fine-grained cognitive impairment detection. The dataset comprises strictly synchronized audio, visual, and linguistic data continuously collected from 574 participants during a naturalistic picture description task. To establish highly reliable diagnostic ground truth, expert clinical neuropsychologists conducted exhaustive evaluations, stratifying participants into distinct cognitive groups through a comprehensive clinical consensus. Consequently, CogPic stands as the largest, most modality-rich, and most meticulously evaluated dataset of its kind to date. By conducting extensive benchmark experiments on the CogPic dataset, we establish an exceptionally robust, unbiased, and clinically generalizable foundation to propel future multimedia research in automated cognitive health assessment. Detailed information and access application procedures for our CogPic database are available at https://cogpic.github.io/.

Qiuqiang Kong, Yin Cao, Haohe Liu et al.

Deep neural network based methods have been successfully applied to music source separation. They typically learn a mapping from a mixture spectrogram to a set of source spectrograms, all with magnitudes only. This approach has several limitations: 1) its incorrect phase reconstruction degrades the performance, 2) it limits the magnitude of masks between 0 and 1 while we observe that 22% of time-frequency bins have ideal ratio mask values of over~1 in a popular dataset, MUSDB18, 3) its potential on very deep architectures is under-explored. Our proposed system is designed to overcome these. First, we propose to estimate phases by estimating complex ideal ratio masks (cIRMs) where we decouple the estimation of cIRMs into magnitude and phase estimations. Second, we extend the separation method to effectively allow the magnitude of the mask to be larger than 1. Finally, we propose a residual UNet architecture with up to 143 layers. Our proposed system achieves a state-of-the-art MSS result on the MUSDB18 dataset, especially, a SDR of 8.98~dB on vocals, outperforming the previous best performance of 7.24~dB. The source code is available at: https://github.com/bytedance/music_source_separation

Qiuqiang Kong, Yin Cao, Turab Iqbal et al.

Audio pattern recognition is an important research topic in the machine learning area, and includes several tasks such as audio tagging, acoustic scene classification, music classification, speech emotion classification and sound event detection. Recently, neural networks have been applied to tackle audio pattern recognition problems. However, previous systems are built on specific datasets with limited durations. Recently, in computer vision and natural language processing, systems pretrained on large-scale datasets have generalized well to several tasks. However, there is limited research on pretraining systems on large-scale datasets for audio pattern recognition. In this paper, we propose pretrained audio neural networks (PANNs) trained on the large-scale AudioSet dataset. These PANNs are transferred to other audio related tasks. We investigate the performance and computational complexity of PANNs modeled by a variety of convolutional neural networks. We propose an architecture called Wavegram-Logmel-CNN using both log-mel spectrogram and waveform as input feature. Our best PANN system achieves a state-of-the-art mean average precision (mAP) of 0.439 on AudioSet tagging, outperforming the best previous system of 0.392. We transfer PANNs to six audio pattern recognition tasks, and demonstrate state-of-the-art performance in several of those tasks. We have released the source code and pretrained models of PANNs: https://github.com/qiuqiangkong/audioset_tagging_cnn.

Fang Kang, Yin Cao, Haoyu Chen

Recent studies in speech-driven talking face generation achieve promising results, but their reliance on fixed-driven speech limits further applications (e.g., face-voice mismatch). Thus, we extend the task to a more challenging setting: given a face image and text to speak, generating both talking face animation and its corresponding speeches. Accordingly, we propose a novel framework, Face2VoiceSync, with several novel contributions: 1) Voice-Face Alignment, ensuring generated voices match facial appearance; 2) Diversity \& Manipulation, enabling generated voice control over paralinguistic features space; 3) Efficient Training, using a lightweight VAE to bridge visual and audio large-pretrained models, with significantly fewer trainable parameters than existing methods; 4) New Evaluation Metric, fairly assessing the diversity and identity consistency. Experiments show Face2VoiceSync achieves both visual and audio state-of-the-art performances on a single 40GB GPU.

Turab Iqbal, Yin Cao, Andrew Bailey et al.

The availability of audio data on sound sharing platforms such as Freesound gives users access to large amounts of annotated audio. Utilising such data for training is becoming increasingly popular, but the problem of label noise that is often prevalent in such datasets requires further investigation. This paper introduces ARCA23K, an Automatically Retrieved and Curated Audio dataset comprised of over 23000 labelled Freesound clips. Unlike past datasets such as FSDKaggle2018 and FSDnoisy18K, ARCA23K facilitates the study of label noise in a more controlled manner. We describe the entire process of creating the dataset such that it is fully reproducible, meaning researchers can extend our work with little effort. We show that the majority of labelling errors in ARCA23K are due to out-of-vocabulary audio clips, and we refer to this type of label noise as open-set label noise. Experiments are carried out in which we study the impact of label noise in terms of classification performance and representation learning.

Yin Cao, Turab Iqbal, Qiuqiang Kong et al.

Polyphonic sound event localization and detection (SELD), which jointly performs sound event detection (SED) and direction-of-arrival (DoA) estimation, detects the type and occurrence time of sound events as well as their corresponding DoA angles simultaneously. We study the SELD task from a multi-task learning perspective. Two open problems are addressed in this paper. Firstly, to detect overlapping sound events of the same type but with different DoAs, we propose to use a trackwise output format and solve the accompanying track permutation problem with permutation-invariant training. Multi-head self-attention is further used to separate tracks. Secondly, a previous finding is that, by using hard parameter-sharing, SELD suffers from a performance loss compared with learning the subtasks separately. This is solved by a soft parameter-sharing scheme. We term the proposed method as Event Independent Network V2 (EINV2), which is an improved version of our previously-proposed method and an end-to-end network for SELD. We show that our proposed EINV2 for joint SED and DoA estimation outperforms previous methods by a large margin, and has comparable performance to state-of-the-art ensemble models.

Yin Cao, Turab Iqbal, Qiuqiang Kong et al.

Polyphonic sound event localization and detection is not only detecting what sound events are happening but localizing corresponding sound sources. This series of tasks was first introduced in DCASE 2019 Task 3. In 2020, the sound event localization and detection task introduces additional challenges in moving sound sources and overlapping-event cases, which include two events of the same type with two different direction-of-arrival (DoA) angles. In this paper, a novel event-independent network for polyphonic sound event localization and detection is proposed. Unlike the two-stage method we proposed in DCASE 2019 Task 3, this new network is fully end-to-end. Inputs to the network are first-order Ambisonics (FOA) time-domain signals, which are then fed into a 1-D convolutional layer to extract acoustic features. The network is then split into two parallel branches. The first branch is for sound event detection (SED), and the second branch is for DoA estimation. There are three types of predictions from the network, SED predictions, DoA predictions, and event activity detection (EAD) predictions that are used to combine the SED and DoA features for on-set and off-set estimation. All of these predictions have the format of two tracks indicating that there are at most two overlapping events. Within each track, there could be at most one event happening. This architecture introduces a problem of track permutation. To address this problem, a frame-level permutation invariant training method is used. Experimental results show that the proposed method can detect polyphonic sound events and their corresponding DoAs. Its performance on the Task 3 dataset is greatly increased as compared with that of the baseline method.

Turab Iqbal, Yin Cao, Qiuqiang Kong et al.

In supervised machine learning, the assumption that training data is labelled correctly is not always satisfied. In this paper, we investigate an instance of labelling error for classification tasks in which the dataset is corrupted with out-of-distribution (OOD) instances: data that does not belong to any of the target classes, but is labelled as such. We show that detecting and relabelling certain OOD instances, rather than discarding them, can have a positive effect on learning. The proposed method uses an auxiliary classifier, trained on data that is known to be in-distribution, for detection and relabelling. The amount of data required for this is shown to be small. Experiments are carried out on the FSDnoisy18k audio dataset, where OOD instances are very prevalent. The proposed method is shown to improve the performance of convolutional neural networks by a significant margin. Comparisons with other noise-robust techniques are similarly encouraging.

Qiuqiang Kong, Yuxuan Wang, Xuchen Song et al.

Source separation is the task to separate an audio recording into individual sound sources. Source separation is fundamental for computational auditory scene analysis. Previous work on source separation has focused on separating particular sound classes such as speech and music. Many of previous work require mixture and clean source pairs for training. In this work, we propose a source separation framework trained with weakly labelled data. Weakly labelled data only contains the tags of an audio clip, without the occurrence time of sound events. We first train a sound event detection system with AudioSet. The trained sound event detection system is used to detect segments that are mostly like to contain a target sound event. Then a regression is learnt from a mixture of two randomly selected segments to a target segment conditioned on the audio tagging prediction of the target segment. Our proposed system can separate 527 kinds of sound classes from AudioSet within a single system. A U-Net is adopted for the separation system and achieves an average SDR of 5.67 dB over 527 sound classes in AudioSet.

Yin Cao, Qiuqiang Kong, Turab Iqbal et al.

Sound event detection (SED) and localization refer to recognizing sound events and estimating their spatial and temporal locations. Using neural networks has become the prevailing method for SED. In the area of sound localization, which is usually performed by estimating the direction of arrival (DOA), learning-based methods have recently been developed. In this paper, it is experimentally shown that the trained SED model is able to contribute to the direction of arrival estimation (DOAE). However, joint training of SED and DOAE degrades the performance of both. Based on these results, a two-stage polyphonic sound event detection and localization method is proposed. The method learns SED first, after which the learned feature layers are transferred for DOAE. It then uses the SED ground truth as a mask to train DOAE. The proposed method is evaluated on the DCASE 2019 Task 3 dataset, which contains different overlapping sound events in different environments. Experimental results show that the proposed method is able to improve the performance of both SED and DOAE, and also performs significantly better than the baseline method.

Qiuqiang Kong, Yin Cao, Turab Iqbal et al.

The Detection and Classification of Acoustic Scenes and Events (DCASE) 2019 challenge focuses on audio tagging, sound event detection and spatial localisation. DCASE 2019 consists of five tasks: 1) acoustic scene classification, 2) audio tagging with noisy labels and minimal supervision, 3) sound event localisation and detection, 4) sound event detection in domestic environments, and 5) urban sound tagging. In this paper, we propose generic cross-task baseline systems based on convolutional neural networks (CNNs). The motivation is to investigate the performance of a variety of models across several tasks without exploiting the specific characteristics of the tasks. We looked at CNNs with 5, 9, and 13 layers, and found that the optimal architecture is task-dependent. For the systems we considered, we found that the 9-layer CNN with average pooling is a good model for a majority of the DCASE 2019 tasks.

Qiuqiang Kong, Yin Cao, Turab Iqbal et al.

The Detection and Classification of Acoustic Scenes and Events (DCASE) 2019 challenge focuses on audio tagging, sound event detection and spatial localisation. DCASE 2019 consists of five tasks: 1) acoustic scene classification, 2) audio tagging with noisy labels and minimal supervision, 3) sound event localisation and detection, 4) sound event detection in domestic environments, and 5) urban sound tagging. In this paper, we propose generic cross-task baseline systems based on convolutional neural networks (CNNs). The motivation is to investigate the performance of a variety of models across several audio recognition tasks without exploiting the specific characteristics of the tasks. We looked at CNNs with 5, 9, and 13 layers, and found that the optimal architecture is task-dependent. For the systems we considered, we found that the 9-layer CNN with average pooling after convolutional layers is a good model for a majority of the DCASE 2019 tasks.

Yin Cao, Bao Wang, Kelin Xia et al.

The matched interface and boundary (MIB) method has a proven ability for delivering the second order accuracy in handling elliptic interface problems with arbitrarily complex interface geometries. However, its collocation formulation requires relatively high solution regularity. Finite volume method (FVM) has its merit in dealing with conservation law problems and its integral formulation works well with relatively low solution regularity. We propose an MIB-FVM to take the advantages of both MIB and FVM for solving elliptic interface problems. We construct the proposed method on Cartesian meshes with vertex-centered control volumes. A large number of numerical experiments are designed to validate the present method in both two dimensional (2D) and three dimensional (3D) domains. It is found that the proposed MIB-FVM achieves the second order convergence for elliptic interface problems with complex interface geometries in both $L_{\infty}$ and $L_2$ norms.