Zhiyao Duan

Samuele Cornell, Jordan Darefsky, Zhiyao Duan et al. · cmu

Currently, a common approach in many speech processing tasks is to leverage large scale pre-trained models by fine-tuning them on in-domain data for a particular application. Yet obtaining even a small amount of such data can be problematic, especially for sensitive domains and conversational speech scenarios, due to both privacy issues and annotation costs. To address this, synthetic data generation using single speaker datasets has been employed. Yet, for multi-speaker cases, such an approach often requires extensive manual effort and is prone to domain mismatches. In this work, we propose a synthetic data generation pipeline for multi-speaker conversational ASR, leveraging a large language model (LLM) for content creation and a conversational multi-speaker text-to-speech (TTS) model for speech synthesis. We conduct evaluation by fine-tuning the Whisper ASR model for telephone and distant conversational speech settings, using both in-domain data and generated synthetic data. Our results show that the proposed method is able to significantly outperform classical multi-speaker generation approaches that use external, non-conversational speech datasets.

Abudukelimu Wuerkaixi, You Zhang, Zhiyao Duan et al.

Active speaker detection (ASD) systems are important modules for analyzing multi-talker conversations. They aim to detect which speakers or none are talking in a visual scene at any given time. Existing research on ASD does not agree on the definition of active speakers. We clarify the definition in this work and require synchronization between the audio and visual speaking activities. This clarification of definition is motivated by our extensive experiments, through which we discover that existing ASD methods fail in modeling the audio-visual synchronization and often classify unsynchronized videos as active speaking. To address this problem, we propose a cross-modal contrastive learning strategy and apply positional encoding in attention modules for supervised ASD models to leverage the synchronization cue. Experimental results suggest that our model can successfully detect unsynchronized speaking as not speaking, addressing the limitation of current models.

Siwen Ding, You Zhang, Zhiyao Duan

Voice anti-spoofing systems are crucial auxiliaries for automatic speaker verification (ASV) systems. A major challenge is caused by unseen attacks empowered by advanced speech synthesis technologies. Our previous research on one-class learning has improved the generalization ability to unseen attacks by compacting the bona fide speech in the embedding space. However, such compactness lacks consideration of the diversity of speakers. In this work, we propose speaker attractor multi-center one-class learning (SAMO), which clusters bona fide speech around a number of speaker attractors and pushes away spoofing attacks from all the attractors in a high-dimensional embedding space. For training, we propose an algorithm for the co-optimization of bona fide speech clustering and bona fide/spoof classification. For inference, we propose strategies to enable anti-spoofing for speakers without enrollment. Our proposed system outperforms existing state-of-the-art single systems with a relative improvement of 38% on equal error rate (EER) on the ASVspoof2019 LA evaluation set.

Frank Cwitkowitz, Jonathan Driedger, Zhiyao Duan

Guitar tablature transcription is an important but understudied problem within the field of music information retrieval. Traditional signal processing approaches offer only limited performance on the task, and there is little acoustic data with transcription labels for training machine learning models. However, guitar transcription labels alone are more widely available in the form of tablature, which is commonly shared among guitarists online. In this work, a collection of symbolic tablature is leveraged to estimate the pairwise likelihood of notes on the guitar. The output layer of a baseline tablature transcription model is reformulated, such that an inhibition loss can be incorporated to discourage the co-activation of unlikely note pairs. This naturally enforces playability constraints for guitar, and yields tablature which is more consistent with the symbolic data used to estimate pairwise likelihoods. With this methodology, we show that symbolic tablature can be used to shape the distribution of a tablature transcription model's predictions, even when little acoustic data is available.

Geoffroy Peeters, Zafar Rafii, Magdalena Fuentes et al.

In this paper, we trace the evolution of Music Information Retrieval (MIR) over the past 25 years. While MIR gathers all kinds of research related to music informatics, a large part of it focuses on signal processing techniques for music data, fostering a close relationship with the IEEE Audio and Acoustic Signal Processing Technical Commitee. In this paper, we reflect the main research achievements of MIR along the three EDICS related to music analysis, processing and generation. We then review a set of successful practices that fuel the rapid development of MIR research. One practice is the annual research benchmark, the Music Information Retrieval Evaluation eXchange, where participants compete on a set of research tasks. Another practice is the pursuit of reproducible and open research. The active engagement with industry research and products is another key factor for achieving large societal impacts and motivating younger generations of students to join the field. Last but not the least, the commitment to diversity, equity and inclusion ensures MIR to be a vibrant and open community where various ideas, methodologies, and career pathways collide. We finish by providing future challenges MIR will have to face.

Yongyi Zang, You Zhang, Mojtaba Heydari et al.

The rise of singing voice synthesis presents critical challenges to artists and industry stakeholders over unauthorized voice usage. Unlike synthesized speech, synthesized singing voices are typically released in songs containing strong background music that may hide synthesis artifacts. Additionally, singing voices present different acoustic and linguistic characteristics from speech utterances. These unique properties make singing voice deepfake detection a relevant but significantly different problem from synthetic speech detection. In this work, we propose the singing voice deepfake detection task. We first present SingFake, the first curated in-the-wild dataset consisting of 28.93 hours of bonafide and 29.40 hours of deepfake song clips in five languages from 40 singers. We provide a train/validation/test split where the test sets include various scenarios. We then use SingFake to evaluate four state-of-the-art speech countermeasure systems trained on speech utterances. We find these systems lag significantly behind their performance on speech test data. When trained on SingFake, either using separated vocal tracks or song mixtures, these systems show substantial improvement. However, our evaluations also identify challenges associated with unseen singers, communication codecs, languages, and musical contexts, calling for dedicated research into singing voice deepfake detection. The SingFake dataset and related resources are available at https://www.singfake.org/.

Ge Zhu, Yujia Yan, Juan-Pablo Caceres et al.

Non-linguistic filler words, such as "uh" or "um", are prevalent in spontaneous speech and serve as indicators for expressing hesitation or uncertainty. Previous works for detecting certain non-linguistic filler words are highly dependent on transcriptions from a well-established commercial automatic speech recognition (ASR) system. However, certain ASR systems are not universally accessible from many aspects, e.g., budget, target languages, and computational power. In this work, we investigate filler word detection system that does not depend on ASR systems. We show that, by using the structured state space sequence model (S4) and neural semi-Markov conditional random fields (semi-CRFs), we achieve an absolute F1 improvement of 6.4% (segment level) and 3.1% (event level) on the PodcastFillers dataset. We also conduct a qualitative analysis on the detected results to analyze the limitations of our proposed system.

Kyungbok Lee, You Zhang, Zhiyao Duan

This paper addresses the challenge of developing a robust audio-visual deepfake detection model. In practical use cases, new generation algorithms are continually emerging, and these algorithms are not encountered during the development of detection methods. This calls for the generalization ability of the method. Additionally, to ensure the credibility of detection methods, it is beneficial for the model to interpret which cues from the video indicate it is fake. Motivated by these considerations, we then propose a multi-stream fusion approach with one-class learning as a representation-level regularization technique. We study the generalization problem of audio-visual deepfake detection by creating a new benchmark by extending and re-splitting the existing FakeAVCeleb dataset. The benchmark contains four categories of fake videos (Real Audio-Fake Visual, Fake Audio-Fake Visual, Fake Audio-Real Visual, and Unsynchronized videos). The experimental results demonstrate that our approach surpasses the previous models by a large margin. Furthermore, our proposed framework offers interpretability, indicating which modality the model identifies as more likely to be fake. The source code is released at https://github.com/bok-bok/MSOC.

Zehua Kcriss Li, Meiying Melissa Chen, Yi Zhong et al.

Expressive speech synthesis aims to generate speech that captures a wide range of para-linguistic features, including emotion and articulation, though current research primarily emphasizes emotional aspects over the nuanced articulatory features mastered by professional voice actors. Inspired by this, we explore expressive speech synthesis through the lens of articulatory phonetics. Specifically, we define a framework with three dimensions: Glottalization, Tenseness, and Resonance (GTR), to guide the synthesis at the voice production level. With this framework, we record a high-quality speech dataset named GTR-Voice, featuring 20 Chinese sentences articulated by a professional voice actor across 125 distinct GTR combinations. We verify the framework and GTR annotations through automatic classification and listening tests, and demonstrate precise controllability along the GTR dimensions on two fine-tuned expressive TTS models. We open-source the dataset and TTS models.

Kyungbok Lee, You Zhang, Zhiyao Duan

The goal of Audio-Visual Segmentation (AVS) is to localize and segment the sounding source objects from video frames. Research on AVS suffers from data scarcity due to the high cost of fine-grained manual annotations. Recent works attempt to overcome the challenge of limited data by leveraging the vision foundation model, Segment Anything Model (SAM), prompting it with audio to enhance its ability to segment sounding source objects. While this approach alleviates the model's burden on understanding visual modality by utilizing knowledge of pre-trained SAM, it does not address the fundamental challenge of learning audio-visual correspondence with limited data. To address this limitation, we propose \textbf{AV2T-SAM}, a novel framework that bridges audio features with the text embedding space of pre-trained text-prompted SAM. Our method leverages multimodal correspondence learned from rich text-image paired datasets to enhance audio-visual alignment. Furthermore, we introduce a novel feature, $\mathbf{\textit{\textbf{f}}_{CLIP} \odot \textit{\textbf{f}}_{CLAP}}$, which emphasizes shared semantics of audio and visual modalities while filtering irrelevant noise. Our approach outperforms existing methods on the AVSBench dataset by effectively utilizing pre-trained segmentation models and cross-modal semantic alignment. The source code is released at https://github.com/bok-bok/AV2T-SAM.

Haixi Zhang, Aiyinsi Zuo, Zirui Li et al.

Amidst the rapid advancement of camera-based autonomous driving technology, effectiveness is often prioritized with limited attention to computational efficiency. To address this issue, this paper introduces LRHPerception, a real-time monocular perception package for autonomous driving that uses single-view camera video to interpret the surrounding environment. The proposed system combines the computational efficiency of end-to-end learning with the rich representational detail of local mapping methodologies. With significant improvements in object tracking and prediction, road segmentation, and depth estimation integrated into a unified framework, LRHPerception processes monocular image data into a five-channel tensor consisting of RGB, road segmentation, and pixel-level depth estimation, augmented with object detection and trajectory prediction. Experimental results demonstrate strong performance, achieving real-time processing at 29 FPS on a single GPU, representing a 555% speedup over the fastest mapping-based approach.

Sefik Emre Eskimez, Kenneth Imade, Na Yang et al. · apple-ml

The fact that emotions play a vital role in social interactions, along with the demand for novel human-computer interaction applications, have led to the development of a number of automatic emotion classification systems. However, it is still debatable whether the performance of such systems can compare with human coders. To address this issue, in this study, we present a comprehensive comparison in a speech-based emotion classification task between 138 Amazon Mechanical Turk workers (Turkers) and a state-of-the-art automatic computer system. The comparison includes classifying speech utterances into six emotions (happy, neutral, sad, anger, disgust and fear), into three arousal classes (active, passive, and neutral), and into three valence classes (positive, negative, and neutral). The results show that the computer system outperforms the naive Turkers in almost all cases. Furthermore, the computer system can increase the classification accuracy by rejecting to classify utterances for which it is not confident, while the Turkers do not show a significantly higher classification accuracy on their confident utterances versus unconfident ones.

Yujia Yan, Zhiyao Duan

The neural semi-Markov Conditional Random Field (semi-CRF) framework has demonstrated promise for event-based piano transcription. In this framework, all events (notes or pedals) are represented as closed time intervals tied to specific event types. The neural semi-CRF approach requires an interval scoring matrix that assigns a score for every candidate interval. However, designing an efficient and expressive architecture for scoring intervals is not trivial. This paper introduces a simple method for scoring intervals using scaled inner product operations that resemble how attention scoring is done in transformers. We show theoretically that, due to the special structure from encoding the non-overlapping intervals, under a mild condition, the inner product operations are expressive enough to represent an ideal scoring matrix that can yield the correct transcription result. We then demonstrate that an encoder-only structured non-hierarchical transformer backbone, operating only on a low-time-resolution feature map, is capable of transcribing piano notes and pedals with high accuracy and time precision. The experiment shows that our approach achieves the new state-of-the-art performance across all subtasks in terms of the F1 measure on the Maestro dataset.

You Zhang, Yongyi Zang, Jiatong Shi et al.

The rapid advancement of AI-generated singing voices, which now closely mimic natural human singing and align seamlessly with musical scores, has led to heightened concerns for artists and the music industry. Unlike spoken voice, singing voice presents unique challenges due to its musical nature and the presence of strong background music, making singing voice deepfake detection (SVDD) a specialized field requiring focused attention. To promote SVDD research, we recently proposed the "SVDD Challenge," the very first research challenge focusing on SVDD for lab-controlled and in-the-wild bonafide and deepfake singing voice recordings. The challenge will be held in conjunction with the 2024 IEEE Spoken Language Technology Workshop (SLT 2024).

Frank Cwitkowitz, Zhiyao Duan

Multi-pitch estimation is a decades-long research problem involving the detection of pitch activity associated with concurrent musical events within multi-instrument mixtures. Supervised learning techniques have demonstrated solid performance on more narrow characterizations of the task, but suffer from limitations concerning the shortage of large-scale and diverse polyphonic music datasets with multi-pitch annotations. We present a suite of self-supervised learning objectives for multi-pitch estimation, which encourage the concentration of support around harmonics, invariance to timbral transformations, and equivariance to geometric transformations. These objectives are sufficient to train an entirely convolutional autoencoder to produce multi-pitch salience-grams directly, without any fine-tuning. Despite training exclusively on a collection of synthetic single-note audio samples, our fully self-supervised framework generalizes to polyphonic music mixtures, and achieves performance comparable to supervised models trained on conventional multi-pitch datasets.

Christodoulos Benetatos, Frank Cwitkowitz, Nathan Pruyne et al.

HARP 2.0 brings deep learning models to digital audio workstation (DAW) software through hosted, asynchronous, remote processing, allowing users to route audio from a plug-in interface through any compatible Gradio endpoint to perform arbitrary transformations. HARP renders endpoint-defined controls and processed audio in-plugin, meaning users can explore a variety of cutting-edge deep learning models without ever leaving the DAW. In the 2.0 release we introduce support for MIDI-based models and audio/MIDI labeling models, provide a streamlined pyharp Python API for model developers, and implement numerous interface and stability improvements. Through this work, we hope to bridge the gap between model developers and creatives, improving access to deep learning models by seamlessly integrating them into DAW workflows.

Yu Zhang, Baotong Tian, Zhiyao Duan

Zero-shot online voice conversion (VC) holds significant promise for real-time communications and entertainment. However, current VC models struggle to preserve semantic fidelity under real-time constraints, deliver natural-sounding conversions, and adapt effectively to unseen speaker characteristics. To address these challenges, we introduce Conan, a chunkwise online zero-shot voice conversion model that preserves the content of the source while matching the voice timbre and styles of reference speech. Conan comprises three core components: 1) a Stream Content Extractor that leverages Emformer for low-latency streaming content encoding; 2) an Adaptive Style Encoder that extracts fine-grained stylistic features from reference speech for enhanced style adaptation; 3) a Causal Shuffle Vocoder that implements a fully causal HiFiGAN using a pixel-shuffle mechanism. Experimental evaluations demonstrate that Conan outperforms baseline models in subjective and objective metrics. Audio samples can be found at https://aaronz345.github.io/ConanDemo.

Frank Cwitkowitz, Zhiyao Duan

Multi-Pitch Estimation (MPE) continues to be a sought after capability of Music Information Retrieval (MIR) systems, and is critical for many applications and downstream tasks involving pitch, including music transcription. However, existing methods are largely based on supervised learning, and there are significant challenges in collecting annotated data for the task. Recently, self-supervised techniques exploiting intrinsic properties of pitch and harmonic signals have shown promise for both monophonic and polyphonic pitch estimation, but these still remain inferior to supervised methods. In this work, we extend the classic supervised MPE paradigm by incorporating several self-supervised objectives based on pitch-invariant and pitch-equivariant properties. This joint training results in a substantial improvement under closed training conditions, which naturally suggests that applying the same objectives to a broader collection of data will yield further improvements. However, in doing so we uncover a phenomenon whereby our model simultaneously overfits to the supervised data while degenerating on data used for self-supervision only. We demonstrate and investigate this and offer our insights on the underlying problem.

You Zhang, Ge Zhu, Zhiyao Duan

The performance of automatic speaker verification (ASV) systems could be degraded by voice spoofing attacks. Most existing works aimed to develop standalone spoofing countermeasure (CM) systems. Relatively little work targeted at developing an integrated spoofing aware speaker verification (SASV) system. In the recent SASV challenge, the organizers encourage the development of such integration by releasing official protocols and baselines. In this paper, we build a probabilistic framework for fusing the ASV and CM subsystem scores. We further propose fusion strategies for direct inference and fine-tuning to predict the SASV score based on the framework. Surprisingly, these strategies significantly improve the SASV equal error rate (EER) from 19.31% of the baseline to 1.53% on the official evaluation trials of the SASV challenge. We verify the effectiveness of our proposed components through ablation studies and provide insights with score distribution analysis.

Mojtaba Heydari, Matthew McCallum, Andreas Ehmann et al.

Inferring music time structures has a broad range of applications in music production, processing and analysis. Scholars have proposed various methods to analyze different aspects of time structures, such as beat, downbeat, tempo and meter. Many state-of-the-art (SOFA) methods, however, are computationally expensive. This makes them inapplicable in real-world industrial settings where the scale of the music collections can be millions. This paper proposes a new state space and a semi-Markov model for music time structure analysis. The proposed approach turns the commonly used 2D state spaces into a 1D model through a jump-back reward strategy. It reduces the state spaces size drastically. We then utilize the proposed method for causal, joint beat, downbeat, tempo, and meter tracking, and compare it against several previous methods. The proposed method delivers similar performance with the SOFA joint causal models with a much smaller state space and a more than 30 times speedup.

Ge Zhu, Frank Cwitkowitz, Zhiyao Duan

In this paper, we conduct a cross-dataset study on parametric and non-parametric raw-waveform based speaker embeddings through speaker verification experiments. In general, we observe a more significant performance degradation of these raw-waveform systems compared to spectral based systems. We then propose two strategies to improve the performance of raw-waveform based systems on cross-dataset tests. The first strategy is to change the real-valued filters into analytic filters to ensure shift-invariance. The second strategy is to apply variational dropout to non-parametric filters to prevent them from overfitting irrelevant nuance features.

Frank Cwitkowitz, Mojtaba Heydari, Zhiyao Duan

In recent years, filterbank learning has become an increasingly popular strategy for various audio-related machine learning tasks. This is partly due to its ability to discover task-specific audio characteristics which can be leveraged in downstream processing. It is also a natural extension of the nearly ubiquitous deep learning methods employed to tackle a diverse array of audio applications. In this work, several variations of a frontend filterbank learning module are investigated for piano transcription, a challenging low-level music information retrieval task. We build upon a standard piano transcription model, modifying only the feature extraction stage. The filterbank module is designed such that its complex filters are unconstrained 1D convolutional kernels with long receptive fields. Additional variations employ the Hilbert transform to render the filters intrinsically analytic and apply variational dropout to promote filterbank sparsity. Transcription results are compared across all experiments, and we offer visualization and analysis of the filterbanks.

Mojtaba Heydari, Frank Cwitkowitz, Zhiyao Duan

The online estimation of rhythmic information, such as beat positions, downbeat positions, and meter, is critical for many real-time music applications. Musical rhythm comprises complex hierarchical relationships across time, rendering its analysis intrinsically challenging and at times subjective. Furthermore, systems which attempt to estimate rhythmic information in real-time must be causal and must produce estimates quickly and efficiently. In this work, we introduce an online system for joint beat, downbeat, and meter tracking, which utilizes causal convolutional and recurrent layers, followed by a pair of sequential Monte Carlo particle filters applied during inference. The proposed system does not need to be primed with a time signature in order to perform downbeat tracking, and is instead able to estimate meter and adjust the predictions over time. Additionally, we propose an information gate strategy to significantly decrease the computational cost of particle filtering during the inference step, making the system much faster than previous sampling-based methods. Experiments on the GTZAN dataset, which is unseen during training, show that the system outperforms various online beat and downbeat tracking systems and achieves comparable performance to a baseline offline joint method.

Xinhui Chen, You Zhang, Ge Zhu et al.

In this paper, we present UR-AIR system submission to the logical access (LA) and the speech deepfake (DF) tracks of the ASVspoof 2021 Challenge. The LA and DF tasks focus on synthetic speech detection (SSD), i.e. detecting text-to-speech and voice conversion as spoofing attacks. Different from previous ASVspoof challenges, the LA task this year presents codec and transmission channel variability, while the new task DF presents general audio compression. Built upon our previous research work on improving the robustness of the SSD systems to channel effects, we propose a channel-robust synthetic speech detection system for the challenge. To mitigate the channel variability issue, we use an acoustic simulator to apply transmission codec, compression codec, and convolutional impulse responses to augmenting the original datasets. For the neural network backbone, we propose to use Emphasized Channel Attention, Propagation and Aggregation Time Delay Neural Networks (ECAPA-TDNN) as our primary model. We also incorporate one-class learning with channel-robust training strategies to further learn a channel-invariant speech representation. Our submission achieved EER 20.33% in the DF task; EER 5.46% and min-tDCF 0.3094 in the LA task.

Bochen Li, Yuxuan Wang, Zhiyao Duan

Separating a song into vocal and accompaniment components is an active research topic, and recent years witnessed an increased performance from supervised training using deep learning techniques. We propose to apply the visual information corresponding to the singers' vocal activities to further improve the quality of the separated vocal signals. The video frontend model takes the input of mouth movement and fuses it into the feature embeddings of an audio-based separation framework. To facilitate the network to learn audiovisual correlation of singing activities, we add extra vocal signals irrelevant to the mouth movement to the audio mixture during training. We create two audiovisual singing performance datasets for training and evaluation, respectively, one curated from audition recordings on the Internet, and the other recorded in house. The proposed method outperforms audio-based methods in terms of separation quality on most test recordings. This advantage is especially pronounced when there are backing vocals in the accompaniment, which poses a great challenge for audio-only methods.

You Zhang, Ge Zhu, Fei Jiang et al.

Spoofing countermeasure (CM) systems are critical in speaker verification; they aim to discern spoofing attacks from bona fide speech trials. In practice, however, acoustic condition variability in speech utterances may significantly degrade the performance of CM systems. In this paper, we conduct a cross-dataset study on several state-of-the-art CM systems and observe significant performance degradation compared with their single-dataset performance. Observing differences of average magnitude spectra of bona fide utterances across the datasets, we hypothesize that channel mismatch among these datasets is one important reason. We then verify it by demonstrating a similar degradation of CM systems trained on original but evaluated on channel-shifted data. Finally, we propose several channel robust strategies (data augmentation, multi-task learning, adversarial learning) for CM systems, and observe a significant performance improvement on cross-dataset experiments.

Mojtaba Heydari, Zhiyao Duan

Online beat tracking (OBT) has always been a challenging task. Due to the inaccessibility of future data and the need to make inference in real-time. We propose Do not Look back! (DLB), a novel approach optimized for efficiency when performing OBT. DLB feeds the activations of a unidirectional RNN into an enhanced Monte-Carlo localization model to infer beat positions. Most preexisting OBT methods either apply some offline approaches to a moving window containing past data to make predictions about future beat positions or must be primed with past data at startup to initialize. Meanwhile, our proposed method only uses activation of the current time frame to infer beat positions. As such, without waiting at the beginning to receive a chunk, it provides an immediate beat tracking response, which is critical for many OBT applications. DLB significantly improves beat tracking accuracy over state-of-the-art OBT methods, yielding a similar performance to offline methods.

You Zhang, Fei Jiang, Zhiyao Duan

Human voices can be used to authenticate the identity of the speaker, but the automatic speaker verification (ASV) systems are vulnerable to voice spoofing attacks, such as impersonation, replay, text-to-speech, and voice conversion. Recently, researchers developed anti-spoofing techniques to improve the reliability of ASV systems against spoofing attacks. However, most methods encounter difficulties in detecting unknown attacks in practical use, which often have different statistical distributions from known attacks. Especially, the fast development of synthetic voice spoofing algorithms is generating increasingly powerful attacks, putting the ASV systems at risk of unseen attacks. In this work, we propose an anti-spoofing system to detect unknown synthetic voice spoofing attacks (i.e., text-to-speech or voice conversion) using one-class learning. The key idea is to compact the bona fide speech representation and inject an angular margin to separate the spoofing attacks in the embedding space. Without resorting to any data augmentation methods, our proposed system achieves an equal error rate (EER) of 2.19% on the evaluation set of ASVspoof 2019 Challenge logical access scenario, outperforming all existing single systems (i.e., those without model ensemble).

Ge Zhu, Fei Jiang, Zhiyao Duan

State-of-the-art text-independent speaker verification systems typically use cepstral features or filter bank energies as speech features. Recent studies attempted to extract speaker embeddings directly from raw waveforms and have shown competitive results. In this paper, we propose a novel multi-scale waveform encoder that uses three convolution branches with different time scales to compute speech features from the waveform. These features are then processed by squeeze-and-excitation blocks, a multi-level feature aggregator, and a time delayed neural network (TDNN) to compute speaker embedding. We show that the proposed embeddings outperform existing raw-waveform-based speaker embeddings on speaker verification by a large margin. A further analysis of the learned filters shows that the multi-scale encoder attends to different frequency bands at its different scales while resulting in a more flat overall frequency response than any of the single-scale counterparts.

Runze Su, Fei Tao, Xudong Liu et al.

The applications of short-term user-generated video (UGV), such as Snapchat, and Youtube short-term videos, booms recently, raising lots of multimodal machine learning tasks. Among them, learning the correspondence between audio and visual information from videos is a challenging one. Most previous work of the audio-visual correspondence(AVC) learning only investigated constrained videos or simple settings, which may not fit the application of UGV. In this paper, we proposed new principles for AVC and introduced a new framework to set sight of videos' themes to facilitate AVC learning. We also released the KWAI-AD-AudVis corpus which contained 85432 short advertisement videos (around 913 hours) made by users. We evaluated our proposed approach on this corpus, and it was able to outperform the baseline by 23.15% absolute difference.

Sefik Emre Eskimez, You Zhang, Zhiyao Duan

Visual emotion expression plays an important role in audiovisual speech communication. In this work, we propose a novel approach to rendering visual emotion expression in speech-driven talking face generation. Specifically, we design an end-to-end talking face generation system that takes a speech utterance, a single face image, and a categorical emotion label as input to render a talking face video synchronized with the speech and expressing the conditioned emotion. Objective evaluation on image quality, audiovisual synchronization, and visual emotion expression shows that the proposed system outperforms a state-of-the-art baseline system. Subjective evaluation of visual emotion expression and video realness also demonstrates the superiority of the proposed system. Furthermore, we conduct a human emotion recognition pilot study using generated videos with mismatched emotions among the audio and visual modalities. Results show that humans respond to the visual modality more significantly than the audio modality on this task.

Nan Jiang, Sheng Jin, Zhiyao Duan et al.

This paper presents a deep reinforcement learning algorithm for online accompaniment generation, with potential for real-time interactive human-machine duet improvisation. Different from offline music generation and harmonization, online music accompaniment requires the algorithm to respond to human input and generate the machine counterpart in a sequential order. We cast this as a reinforcement learning problem, where the generation agent learns a policy to generate a musical note (action) based on previously generated context (state). The key of this algorithm is the well-functioning reward model. Instead of defining it using music composition rules, we learn this model from monophonic and polyphonic training data. This model considers the compatibility of the machine-generated note with both the machine-generated context and the human-generated context. Experiments show that this algorithm is able to respond to the human part and generate a melodic, harmonic and diverse machine part. Subjective evaluations on preferences show that the proposed algorithm generates music pieces of higher quality than the baseline method.

Mingrui Yuan, Zhiyao Duan

This paper proposes a deep multi-speaker text-to-speech (TTS) model for spoofing speaker verification (SV) systems. The proposed model employs one network to synthesize time-downsampled mel-spectrograms from text input and another network to convert them to linear-frequency spectrograms, which are further converted to the time domain using the Griffin-Lim algorithm. Both networks are trained separately under the generative adversarial networks (GAN) framework. Spoofing experiments on two state-of-the-art SV systems (i-vectors and Google's GE2E) show that the proposed system can successfully spoof these systems with a high success rate. Spoofing experiments on anti-spoofing systems (i.e., binary classifiers for discriminating real and synthetic speech) also show a high spoof success rate when such anti-spoofing systems' structures are exposed to the proposed TTS system.

Yichi Zhang, Yiting Zhang, Zhiyao Duan

Searching sounds by text labels is often difficult, as text descriptions cannot describe the audio content in detail. Query by vocal imitation bridges such gap and provides a novel way to sound search. Several algorithms for sound search by vocal imitation have been proposed and evaluated in a simulation environment, however, they have not been deployed into a real search engine nor evaluated by real users. This pilot work conducts a subjective study to compare these two approaches to sound search, and tries to answer the question of which approach works better for what kinds of sounds. To do so, we developed two web-based search engines for sound, one by vocal imitation (Vroom!) and the other by text description (TextSearch). We also developed an experimental framework to host these engines to collect statistics of user behaviors and ratings. Results showed that Vroom! received significantly higher search satisfaction ratings than TextSearch did for sound categories that were difficult for subjects to describe by text. Results also showed a better overall ease-of-use rating for Vroom! than TextSearch on the limited sound library in our experiments. These findings suggest advantages of vocal-imitation-based search for sound in practice.

Lele Chen, Ross K. Maddox, Zhiyao Duan et al.

We devise a cascade GAN approach to generate talking face video, which is robust to different face shapes, view angles, facial characteristics, and noisy audio conditions. Instead of learning a direct mapping from audio to video frames, we propose first to transfer audio to high-level structure, i.e., the facial landmarks, and then to generate video frames conditioned on the landmarks. Compared to a direct audio-to-image approach, our cascade approach avoids fitting spurious correlations between audiovisual signals that are irrelevant to the speech content. We, humans, are sensitive to temporal discontinuities and subtle artifacts in video. To avoid those pixel jittering problems and to enforce the network to focus on audiovisual-correlated regions, we propose a novel dynamically adjustable pixel-wise loss with an attention mechanism. Furthermore, to generate a sharper image with well-synchronized facial movements, we propose a novel regression-based discriminator structure, which considers sequence-level information along with frame-level information. Thoughtful experiments on several datasets and real-world samples demonstrate significantly better results obtained by our method than the state-of-the-art methods in both quantitative and qualitative comparisons.

Lele Chen, Zhiheng Li, Ross K. Maddox et al.

Cross-modality generation is an emerging topic that aims to synthesize data in one modality based on information in a different modality. In this paper, we consider a task of such: given an arbitrary audio speech and one lip image of arbitrary target identity, generate synthesized lip movements of the target identity saying the speech. To perform well in this task, it inevitably requires a model to not only consider the retention of target identity, photo-realistic of synthesized images, consistency and smoothness of lip images in a sequence, but more importantly, learn the correlations between audio speech and lip movements. To solve the collective problems, we explore the best modeling of the audio-visual correlations in building and training a lip-movement generator network. Specifically, we devise a method to fuse audio and image embeddings to generate multiple lip images at once and propose a novel correlation loss to synchronize lip changes and speech changes. Our final model utilizes a combination of four losses for a comprehensive consideration in generating lip movements; it is trained in an end-to-end fashion and is robust to lip shapes, view angles and different facial characteristics. Thoughtful experiments on three datasets ranging from lab-recorded to lips in-the-wild show that our model significantly outperforms other state-of-the-art methods extended to this task.

Sefik Emre Eskimez, Ross K Maddox, Chenliang Xu et al.

The presence of a corresponding talking face has been shown to significantly improve speech intelligibility in noisy conditions and for hearing impaired population. In this paper, we present a system that can generate landmark points of a talking face from an acoustic speech in real time. The system uses a long short-term memory (LSTM) network and is trained on frontal videos of 27 different speakers with automatically extracted face landmarks. After training, it can produce talking face landmarks from the acoustic speech of unseen speakers and utterances. The training phase contains three key steps. We first transform landmarks of the first video frame to pin the two eye points into two predefined locations and apply the same transformation on all of the following video frames. We then remove the identity information by transforming the landmarks into a mean face shape across the entire training dataset. Finally, we train an LSTM network that takes the first- and second-order temporal differences of the log-mel spectrogram as input to predict face landmarks in each frame. We evaluate our system using the mean-squared error (MSE) loss of landmarks of lips between predicted and ground-truth landmarks as well as their first- and second-order temporal differences. We further evaluate our system by conducting subjective tests, where the subjects try to distinguish the real and fake videos of talking face landmarks. Both tests show promising results.

Yapeng Tian, Jing Shi, Bochen Li et al.

In this paper, we introduce a novel problem of audio-visual event localization in unconstrained videos. We define an audio-visual event as an event that is both visible and audible in a video segment. We collect an Audio-Visual Event(AVE) dataset to systemically investigate three temporal localization tasks: supervised and weakly-supervised audio-visual event localization, and cross-modality localization. We develop an audio-guided visual attention mechanism to explore audio-visual correlations, propose a dual multimodal residual network (DMRN) to fuse information over the two modalities, and introduce an audio-visual distance learning network to handle the cross-modality localization. Our experiments support the following findings: joint modeling of auditory and visual modalities outperforms independent modeling, the learned attention can capture semantics of sounding objects, temporal alignment is important for audio-visual fusion, the proposed DMRN is effective in fusing audio-visual features, and strong correlations between the two modalities enable cross-modality localization.

Lele Chen, Sudhanshu Srivastava, Zhiyao Duan et al.

Cross-modal audio-visual perception has been a long-lasting topic in psychology and neurology, and various studies have discovered strong correlations in human perception of auditory and visual stimuli. Despite works in computational multimodal modeling, the problem of cross-modal audio-visual generation has not been systematically studied in the literature. In this paper, we make the first attempt to solve this cross-modal generation problem leveraging the power of deep generative adversarial training. Specifically, we use conditional generative adversarial networks to achieve cross-modal audio-visual generation of musical performances. We explore different encoding methods for audio and visual signals, and work on two scenarios: instrument-oriented generation and pose-oriented generation. Being the first to explore this new problem, we compose two new datasets with pairs of images and sounds of musical performances of different instruments. Our experiments using both classification and human evaluations demonstrate that our model has the ability to generate one modality, i.e., audio/visual, from the other modality, i.e., visual/audio, to a good extent. Our experiments on various design choices along with the datasets will facilitate future research in this new problem space.

Bochen Li, Xinzhao Liu, Karthik Dinesh et al.

We introduce a dataset for facilitating audio-visual analysis of music performances. The dataset comprises 44 simple multi-instrument classical music pieces assembled from coordinated but separately recorded performances of individual tracks. For each piece, we provide the musical score in MIDI format, the audio recordings of the individual tracks, the audio and video recording of the assembled mixture, and ground-truth annotation files including frame-level and note-level transcriptions. We describe our methodology for the creation of the dataset, particularly highlighting our approaches for addressing the challenges involved in maintaining synchronization and expressiveness. We demonstrate the high quality of synchronization achieved with our proposed approach by comparing the dataset with existing widely-used music audio datasets. We anticipate that the dataset will be useful for the development and evaluation of existing music information retrieval (MIR) tasks, as well as for novel multi-modal tasks. We benchmark two existing MIR tasks (multi-pitch analysis and score-informed source separation) on the dataset and compare with other existing music audio datasets. Additionally, we consider two novel multi-modal MIR tasks (visually informed multi-pitch analysis and polyphonic vibrato analysis) enabled by the dataset and provide evaluation measures and baseline systems for future comparisons (from our recent work). Finally, we propose several emerging research directions that the dataset enables.