Contrastive language-audio pre-training (CLAP) has addressed audio-language tasks such as audio-text retrieval by aligning audio and text in a common feature space. While CLAP addresses general audio-language tasks, its audio features do not generalize well in audio tasks. In contrast, self-supervised learning (SSL) models learn general-purpose audio features that perform well in diverse audio tasks. We pursue representation learning that can be widely used in audio applications and hypothesize that a method that learns both general audio features and CLAP features should achieve our goal, which we call a general-purpose audio-language representation. To implement our hypothesis, we propose M2D2, a second-generation masked modeling duo (M2D) that combines an SSL M2D and CLAP. M2D2 learns two types of features using two modalities (audio and text) in a two-stage training process. It also utilizes advanced LLM-based sentence embeddings in CLAP training for powerful semantic supervision. In the first stage, M2D2 learns generalizable audio features from M2D and CLAP, where CLAP aligns the features with the fine LLM-based semantic embeddings. In the second stage, it learns CLAP features using the audio features learned from the LLM-based embeddings. Through these pre-training stages, M2D2 should enhance generalizability and performance in its audio and CLAP features. Experiments validated that M2D2 achieves effective general-purpose audio-language representation, highlighted with SOTA fine-tuning mAP of 49.0 for AudioSet, SOTA performance in music tasks, and top-level performance in audio-language tasks.
The scarcity of high-quality and multi-task singing datasets significantly hinders the development of diverse controllable and personalized singing tasks, as existing singing datasets suffer from low quality, limited diversity of languages and singers, absence of multi-technique information and realistic music scores, and poor task suitability. To tackle these problems, we present GTSinger, a large global, multi-technique, free-to-use, high-quality singing corpus with realistic music scores, designed for all singing tasks, along with its benchmarks. Particularly, (1) we collect 80.59 hours of high-quality singing voices, forming the largest recorded singing dataset; (2) 20 professional singers across nine widely spoken languages offer diverse timbres and styles; (3) we provide controlled comparison and phoneme-level annotations of six commonly used singing techniques, helping technique modeling and control; (4) GTSinger offers realistic music scores, assisting real-world musical composition; (5) singing voices are accompanied by manual phoneme-to-audio alignments, global style labels, and 16.16 hours of paired speech for various singing tasks. Moreover, to facilitate the use of GTSinger, we conduct four benchmark experiments: technique-controllable singing voice synthesis, technique recognition, style transfer, and speech-to-singing conversion. The corpus and demos can be found at http://aaronz345.github.io/GTSingerDemo/. We provide the dataset and the code for processing data and conducting benchmarks at https://huggingface.co/datasets/AaronZ345/GTSinger and https://github.com/AaronZ345/GTSinger.
This study introduces self-supervised contrastive learning to acquire feature representations of singing voices. To acquire robust representations in an unsupervised manner, regular self-supervised contrastive learning trains neural networks to make the feature representation of a sample close to those of its computationally transformed versions. Similarly, we employ two transformations—pitch shifting and time stretching—considering the nature of singing voices. Nevertheless, we use them reversely: we train networks to push away representations of the transformed versions. The networks then attempt to discriminate changes in vocal timbres introduced by pitch shifting without time stretching and those in singing expressions introduced by time stretching without pitch shifting. Consequently, the acquired representations become attentive to vocal timbre and singing expression. This was confirmed through a singer identification task, where we trained a classifier to learn the relationship between the feature representations to the corresponding singer labels of 500 singers. As a result, the employed transformations helped the classifier improve the classification accuracy by 9.12% (top-1 accuracy: 63.08%) compared with the case where the feature representations fed to the classifier were acquired without the transformations (top-1 accuracy: 53.96%). Furthermore, the proposed approach can be extended to acquire feature representations attentive to either vocal timbre or singing expression but not to the other by changing how the transformations are incorporated. We particularly explored the characteristics of such vocal timbre- or singing expression-oriented feature representations against song genre, singer gender, and vocal technique, and confirmed that they successfully capture different aspects of singing voices.