Audio-Based Musical Version Identification: Elements and challenges

Yesiler, Furkan; Doras, Guillaume; Bittner, Rachel M.; Tralie, Christopher J.; Serrà, Joan

doi:10.1109/msp.2021.3105941

Cited by 4 publications

(4 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this section, we describe and motivate our design choices. We first present how we extracted our rhythmic and lyrics features, and publicly release our datasets 1 . We then present our metric learning-based VI model.…”

Section: Proposed Methodsmentioning

confidence: 99%

“…Various MIR applications rely on a concept of musical similarity, e.g. music classification [11], music recommendation [12], or VI systems [1], among many others. Musical similarity between two tracks is typically evaluated first deriving an intermediate feature representation from the audio waveform and then computing a distance between feature pairs.…”

Section: Metric Learningmentioning

confidence: 99%

“…Pioneering works showed promising accuracy on small audio datasets, but remained difficult to scale to larger modern audio corpora. The recent introduction of data-driven approaches based on neural networks led to significant progress towards accurate yet scalable VI systems [1].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

And what if two musical versions don't share melody, harmony, rhythm, or lyrics ?

Mathilde¹,

Doras²

2022

Preprint

View full text Add to dashboard Cite

Version identification (VI) has seen substantial progress over the past few years. On the one hand, the introduction of the metric learning paradigm has favored the emergence of scalable yet accurate VI systems. On the other hand, using features focusing on specific aspects of musical pieces, such as melody, harmony, or lyrics, yielded interpretable and promising performances. In this work, we build upon these recent advances and propose a metric learning-based system systematically leveraging four dimensions commonly admitted to convey musical similarity between versions: melodic line, harmonic structure, rhythmic patterns, and lyrics. We describe our deliberately simple model architecture, and we show in particular that an approximated representation of the lyrics is an efficient proxy to discriminate between versions and nonversions. We then describe how these features complement each other and yield new state-of-the-art performances on two publicly available datasets. We finally suggest that a VI system using a combination of melodic, harmonic, rhythmic and lyrics features could theoretically reach the optimal performances obtainable on these datasets.

show abstract

Section: Proposed Methodsmentioning

confidence: 99%

Section: Metric Learningmentioning

confidence: 99%

See 1 more Smart Citation

And what if two musical versions don't share melody, harmony, rhythm, or lyrics ?

Mathilde¹,

Doras²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…We download the data with URLs from second-hand-songs website 1 and use the original setting 2 to split all recordings into train/dev/test sets. In addition to evaluating our method on the test part of SHS100K, we also use Cover80 [3] and DaTacos [18] as test sets for comparison, which have 160 and 15,000 recordings, respectively. Note that we remove the overlap items between SHS100K and DaTacos as suggested by [7].…”

Section: Experiments a Experiments Settingsmentioning

confidence: 99%

Analysis of Social Media Messaging for Health Promotion by the CDC from 2018 to2019 in Guangzhou, China (Preprint)

Liu¹,

Ma²

2019

Preprint

View full text Add to dashboard Cite

BACKGROUND As Internet use becomes increasingly widespread, mobile devices, such as mobile phones, have become the most important channel for many people to obtain information. The number of Chinese mobile phone users has reached 1.28 billion, with more than 90% of usersaccessing social media via the mobile application WeChat. As an increasinglyubiquitous app in China, WeChat had1.04 billion monthly active users worldwide by the first quarter of 2018.[1,2]A recent national survey in China found that one-third of participants regularly read health education articles on WeChat, and 98.53% of participants choose to useWeChat for health information seeking, indicatingthatWeChat is the most popularplatform for health information acquisition in China.[3]“Everyone is the media and everyone has a microphone” in theInternet era,and the rapiddevelopment of communications technologyhas playedan increasingly important role in information dissemination. WeChat is the most popular social media platform in China, and 93% of residents in major Chinese cities are reported to log into WeChatdaily. Social media platforms, particularly WeChat, arewidely utilized by health organizations, and are fast becoming the principal instruments of alternative communication channels fordelivering health messages, conducting disease surveillance, spreading health awareness, and communicatingabout public health issues to the public.WeChat has emerged as a powerful platform with several advantages compared with more traditional communication channels, and has proven to be a cost-effective tool for the dissemination of health messages, capable of reaching minority groupsto improve public health interventions. In recent years, there has been increasing public interest in searching for online health information related to health problems. Evidence from local studies has reportedthat members of thepublic perceive information provided online to be useful and reliable, a perception that leads many peopleto consult with health care practitioners about their health conditions directly through social media. Growing online health-seeking behaviors and the increasing numbers of non-authorized health websites or social media accounts sharingbiasedor inaccurate health information involvingconflicts of interest have made it necessary for health organizations to engage with internet users on social media in a strategic way.[4]To adapt to the reading habits of the public in the Internet era, various institutions have utilized the new forms of media as important channels for government propaganda and the provision of services. WeChat is a new form of media,utilizing mobile phones as terminals, enabling instantaneous and social communication. WeChat is an important platform for audiences to obtain government information. At present, the daily search volume of medical and health problems from online has reached 60 million. A report published by the Ministry of Science Popularization of the Chinese Association of Science and Technology and the Institute of Science Popularization of China titled“Big Data of Internet Popular Science Reading 2017”reported that the popular science articles read most often by citizensin 2017 werethose focused on health and medical treatment. [3] The Guangzhou Centerfor Disease Controland Prevention(GZCDC) i-Health is anofficial WeChat account operated by the GZCDC, officially opened in April 2018. The postspublished via this account are mainly original, focusing on health tips and popular science. Because of differences in functionality between WeChat and other social media platforms, such as Facebook and Twitter, similar studies of social media are not generalizable to WeChat. Thus,specific investigations of the effectiveness of WeChat for health promotion are important.In the current study, posts published between April 2018 and April 2019 were analyzed to understand the public demand for the official CDC WeChat account, and to provide scientific evidence toinform the development of better communication strategies. OBJECTIVE To analyze the datapublished by the official Guangzhou Centerfor Disease Controland Prevention (GZCDC) i-Health WeChat account and explore the factors influencing theeffectsoninformation dissemination, soasto improve the effectiveness of health information dissemination. METHODS Data were collected from the official Guangzhou i-Health WeChat account between 1 April 2018 and 30April 2019. Descriptive analysiswas performed for the basic information regarding the WeChat account and its posts. Multiple-factor logistic regression analysis was used to explore the association between various factors of posts on engagementof followers, and the impact of the WeChat account was examined usingadjusted odds ratios (AOR) and the corresponding 95% confidence intervals (95% CI) for the variables. RESULTS Thetotal numbers of page views, shares, likes, adds to favorites, and comments for213 posts were 1147308, 84671, 10973, 5535 and 1865, respectively, from 187033followers.The engagement of followers peaked on the dissemination date and gradually declined. The main post topicswere health education posts and original posts, and 46.48% of 213posts were reposted by other official WeChat accounts.In a multiple-factor logistic regression model, the number of page viewswas found to be significantly associated withinfectious disease posts (AOR:3.20 95%CI: 1.16–8.81), original posts (AOR:10.20, 95%CI:1.17–89.28), and posts that could be judged from the title (AOR:2.93,95%CI:1.16–8.81). Vaccine-relatedevent posts (AOR:15.78, 95%CI:3.39–73.42), child and adolescent health posts (AOR:17.33, 95%CI:1.50–200.31), environmental health posts (AOR:7.94, 95%CI:1.90–33.21), chronic disease posts(AOR:4.05, 95%CI:1.18–13.45),nutrition and food-borne diseaseposts(AOR:3.91, 95%CI:1.25–12.20), infectious disease posts (AOR:3.86, 95%CI:1.36–10.98),original posts (AOR:10.22, 95%CI:1.06–98.85), posts focused on current events (AOR:3.04, 95%CI=1.36–6.76) and headline posts (AOR:5.48, 95%CI:1.14–26.41) were positively associated with being reposted by other official WeChat accounts. CONCLUSIONS Content is the most important factor in the effectiveness of official WeChat accounts at GZCDC institutions, and should be focused on health tips and popular science articles, while enhancing interactions with the public.

show abstract

Training audio transformers for cover song identification

Zeng,

Lau

2023

J AUDIO SPEECH MUSIC PROC.

View full text Add to dashboard Cite

In the past decades, convolutional neural networks (CNNs) have been commonly adopted in audio perception tasks, which aim to learn latent representations. However, for audio analysis, CNNs may exhibit limitations in effectively modeling temporal contextual information. Analogous to the successes of transformer architecture used in the fields of computer vision and audio classification, to capture long-range global contexts better, we here extend this line of work and propose an Audio Similarity Transformer (ASimT), a convolution-free, purely transformer network-based architecture for learning effective representations of audio signals. Furthermore, we introduce a novel loss MAPLoss, used in tandem with classification loss, to directly enhance the mean average precision. In the experiments, ASimT demonstrates its state-of-the-art performance in cover song identification on public datasets.

show abstract

Audio-Based Musical Version Identification: Elements and challenges

Cited by 4 publications

References 30 publications

And what if two musical versions don't share melody, harmony, rhythm, or lyrics ?

And what if two musical versions don't share melody, harmony, rhythm, or lyrics ?

Analysis of Social Media Messaging for Health Promotion by the CDC from 2018 to2019 in Guangzhou, China (Preprint)

Training audio transformers for cover song identification

Contact Info

Product

Resources

About