This paper describes our submission to the Second Clarity Enhancement Challenge (CEC2), which consists of target speech enhancement for hearing-aid (HA) devices in noisy-reverberant environments with multiple interferers such as music and competing speakers. Our approach builds upon the powerful iterative neural/beamforming enhancement (iNeuBe) framework introduced in our recent work, and this paper extends it for target speaker extraction. We therefore name the proposed approach as iNeuBe-X, where the X stands for extraction. To address the challenges encountered in the CEC2 setting, we introduce four major novelties: (1) we extend the state-of-the-art TF-GridNet model, originally designed for monaural speaker separation, for multi-channel, causal speech enhancement, and large improvements are observed by replacing the TCNDenseNet used in iNeuBe with this new architecture; (2) we leverage a recent dual window size approach with future-frame prediction to ensure that iNueBe-X satisfies the 5 ms constraint on algorithmic latency required by CEC2; (3) we introduce a novel speaker-conditioning branch for TF-GridNet to achieve target speaker extraction; (4) we propose a fine-tuning step, where we compute an additional loss with respect to the target speaker signal compensated with the listener audiogram. Without using external data, on the official development set our best model reaches a hearing-aid speech perception index (HASPI) score of 0.942 and a scale-invariant signal-to-distortion ratio improvement (SI-SDRi) of 18.8 dB. These results are promising given the fact that the CEC2 data is extremely challenging (e.g., on the development set the mixture SI-SDR is -12.3 dB). A demo of our submitted system is available at WAVLab CEC2 demo.
Transformers have drawn attention in the MIR field for their remarkable performance shown in natural language processing and computer vision. However, prior works in the audio processing domain mostly use Transformer as a temporal feature aggregator that acts similar to RNNs. In this paper, we propose SpecTNT, a Transformer-based architecture to model both spectral and temporal sequences of an input time-frequency representation. Specifically, we introduce a novel variant of the Transformer-in-Transformer (TNT) architecture. In each SpecTNT block, a spectral Transformer extracts frequency-related features into the frequency class token (FCT) for each frame. Later, the FCTs are linearly projected and added to the temporal embeddings (TEs), which aggregate useful information from the FCTs. Then, a temporal Transformer processes the TEs to exchange information across the time axis. By stacking the SpecTNT blocks, we build the SpecTNT model to learn the representation for music signals. In experiments, SpecTNT demonstrates state-of-the-art performance in music tagging and vocal melody extraction, and shows competitive performance for chord recognition. The effectiveness of SpecTNT and other design choices are further examined through ablation studies.
Emotion is a complicated notion present in music that is hard to capture even with fine-tuned feature engineering. In this paper, we investigate the utility of state-of-the-art pre-trained deep audio embedding methods to be used in the Music Emotion Recognition (MER) task. Deep audio embedding methods allow us to efficiently capture the high dimensional features into a compact representation. We implement several multi-class classifiers with deep audio embeddings to predict emotion semantics in music. We investigate the effectiveness of L3-Net and VGGish deep audio embedding methods for music emotion inference over four music datasets. The experiments with several classifiers on the task show that the deep audio embedding solutions can improve the performances of the previous baseline MER models. We conclude that deep audio embeddings represent musical emotion semantics for the MER task without expert human engineering.
In this paper, we focus on singing techniques within the scope of music information retrieval research. We investigate how singers use singing techniques using real-world recordings of famous solo singers in Japanese popular music songs (J-POP). First, we built a new dataset of singing techniques. The dataset consists of 168 commercial J-POP songs, and each song is annotated using various singing techniques with timestamps and vocal pitch contours. We also present descriptive statistics of singing techniques on the dataset to clarify what and how often singing techniques appear. We further explored the difficulty of the automatic detection of singing techniques using previously proposed machine learning techniques. In the detection, we also investigate the effectiveness of auxiliary information (i.e., pitch and distribution of label duration), not only providing the baseline. The best result achieves 40.4% at macro-average F-measure on nine-way multi-class detection. We provide the annotation of the dataset and its detail on the appendix website 0 .
One of the advantages of using natural language processing (NLP) technology for music is to fully exploit the embedding based representation learning paradigm that can easily handle classical tasks such as semantic similarity. However, recent researches have revealed the poor performance issue of common baseline methods for semantic similarity in NLP. They show that some simple embedding calibration methods can easily promote the performance of semantic similarity without extra training hence is ready-to-use. Nevertheless, it is still unclear which is the best combination of calibration methods and by how much can we further improve the performance with such methods. Most importantly, previous works are based on auto-encoder Transformer, hence the performance under auto-regressive model for music is unclear. These render the following open questions: does embedding based semantic similarity also apply for auto-regressive music model, does poor baseline issue for semantic similarity also exists, and if so, are there unexplored embedding calibration methods to better promote the performance of music semantic similarity? In this paper, we answer these questions by exploring different combination of embedding calibration under auto-regressive language model for symbolic music. Our results show that music semantic similarity works under auto-regressive model, and also suffers from poor baseline issues like in NLP. Furthermore, we provide optimal combination of embedding calibration that has not been explored in previous researches. Results show that such combination of embedding calibration can greatly improve music semantic similarity without further training tasks.
In recent years, deep learning techniques have been applied to music generation systems with promising results. However, one of the main challenges in this field has been the lack of annotated datasets, making it difficult for models to learn musical forms in compositions. To address this issue, we present TunesFormer, a Transformer-based melody generation system that is trained on a large dataset of 285,449 ABC tunes. By utilizing specific symbols commonly found in ABC notation to indicate section boundaries, TunesFormer can understand and generate melodies with given musical forms based on control codes. Our objective evaluations demonstrate the effectiveness of the control codes in achieving controlled musical forms, and subjective experiments show that the generated melodies are of comparable quality to human compositions. Our results also provide insights into the optimal placement of control codes and their impact on the generated melodies. TunesFormer presents a promising approach for generating melodies with desired musical forms through the use of deep learning techniques.
We propose an audio-to-audio neural network model that learns to denoise old music recordings. Our model internally converts its input into a time-frequency representation by means of a short-time Fourier transform (STFT), and processes the resulting complex spectrogram using a convolutional neural network. The network is trained with both reconstruction and adversarial objectives on a synthetic noisy music dataset, which is created by mixing clean music with real noise samples extracted from quiet segments of old recordings. We evaluate our method quantitatively on held-out test examples of the synthetic dataset, and qualitatively by human rating on samples of actual historical recordings. Our results show that the proposed method is effective in removing noise, while preserving the quality and details of the original music.
There is a long history of an effort made to explore musical elements with the entities and spaces around us, such as musique concr\`ete and ambient music. In the context of computer music and digital art, interactive experiences that concentrate on the surrounding objects and physical spaces have also been designed. In recent years, with the development and popularization of devices, an increasing number of works have been designed in Extended Reality to create such musical experiences. In this paper, we describe MR4MR, a sound installation work that allows users to experience melodies produced from interactions with their surrounding space in the context of Mixed Reality (MR). Using HoloLens, an MR head-mounted display, users can bump virtual objects that emit sound against real objects in their surroundings. Then, by continuously creating a melody following the sound made by the object and re-generating randomly and gradually changing melody using music generation machine learning models, users can feel their ambient melody "reincarnating".
Recent interest in integrated sensing and communications has led to the design of novel signal processing techniques to recover information from an overlaid radar-communications signal. Here, we focus on a spectral coexistence scenario, wherein the channels and transmit signals of both radar and communications systems are unknown to the common receiver. In this dual-blind deconvolution (DBD) problem, the receiver admits a multi-carrier wireless communications signal that is overlaid with the radar signal reflected off multiple targets. The communications and radar channels are represented by continuous-valued range-times or delays corresponding to multiple transmission paths and targets, respectively. Prior works addressed recovery of unknown channels and signals in this ill-posed DBD problem through atomic norm minimization but contingent on individual minimum separation conditions for radar and communications channels. In this paper, we provide an optimal joint separation condition using extremal functions from the Beurling-Selberg interpolation theory. Thereafter, we formulate DBD as a low-rank modified Hankel matrix retrieval and solve it via nuclear norm minimization. We estimate the unknown target and communications parameters from the recovered low-rank matrix using multiple signal classification (MUSIC) method. We show that the joint separation condition also guarantees that the underlying Vandermonde matrix for MUSIC is well-conditioned. Numerical experiments validate our theoretical findings.
HuQin is a family of traditional Chinese bowed string instruments. Playing techniques(PTs) embodied in various playing styles add abundant emotional coloring and aesthetic feelings to HuQin performance. The complex applied techniques make HuQin music a challenging source for fundamental MIR tasks such as pitch analysis, transcription and score-audio alignment. In this paper, we present a multimodal performance dataset of HuQin music that contains audio-visual recordings of 11,992 single PT clips and 57 annotated musical pieces of classical excerpts. We systematically describe the HuQin PT taxonomy based on musicological theory and practical use cases. Then we introduce the dataset creation methodology and highlight the annotation principles featuring PTs. We analyze the statistics in different aspects to demonstrate the variety of PTs played in HuQin subcategories and perform preliminary experiments to show the potential applications of the dataset in various MIR tasks and cross-cultural music studies. Finally, we propose future work to be extended on the dataset.