Jui Shah

Recently, Generative Adversarial Networks (GAN)-based methods have shown remarkable performance for the Voice Conversion and WHiSPer-to-normal SPeeCH (WHSP2SPCH) conversion. One of the key challenges in WHSP2SPCH conversion is the prediction of fundamental frequency (F0). Recently, authors have proposed state-of-the-art method Cycle-Consistent Generative Adversarial Networks (CycleGAN) for WHSP2SPCH conversion. The CycleGAN-based method uses two different models, one for Mel Cepstral Coefficients (MCC) mapping, and another for F0 prediction, where F0 is highly dependent on the pre-trained model of MCC mapping. This leads to additional non-linear noise in predicted $F_0$. To suppress this noise, we propose Cycle-in-Cycle GAN (i.e., CinC-GAN). It is specially designed to increase the effectiveness in F0 prediction without losing the accuracy of MCC mapping. We evaluated the proposed method on a non-parallel setting and analyzed on speaker-specific, and gender-specific tasks. The objective and subjective tests show that CinC-GAN significantly outperforms the CycleGAN. In addition, we analyze the CycleGAN and CinC-GAN for unseen speakers and the results show the clear superiority of CinC-GAN.

Singing voice conversion (SVC) is a task of converting the perception of source speaker’s identity to target speaker without changing lyrics and rhythm. Recent approaches in traditional voice conversion involves the use of the generative models, such as Autoencoders (AE), Variational Autoencoders (VAE), and Generative Adversarial Networks (GANs). However, in case of SVC, GANs are not explored much. The only system that has been proposed in the literature u ses traditional GAN on the parallel data. The parallel data collection for real scenarios (with the same background music) is not feasible at all. Moreover, SVC in the presence of background music is one of the most challenging tasks as it involves the source separation of vocals from the inputs, which will have some amount of noise. Therefore, in this paper, we propose transfer learning, and fine-tuning-based Cycle consistent GAN (CycleGAN) model for non-parallel SVC, where music source separation is done using Deep Atrractor Network (DANet). We designed seven different possible scenarios to identify the best possible combination of transfer learning and fine-tuning. Here, we use more challenging database, MUSDB18, as our primary dataset. And, we use NUS-48E database to pretrain CycleGAN. We perform extensive analysis via objective and subjective measures and report that with a 4.14 MOS score out of 5 for naturalness, the CycleGAN model pretrained on NUS-48E corpus performs the best compared to the other systems we have described in the paper.

This paper proposes a method that optimizes the inverse kinematics needed for the trajectory generation of a 4-DOF (degrees of freedom) robotic manipulator arm to give results in real time. Due to the many-to-one mapping of the angle vector which describes the position of the manipulator joints and to the coordinates of the end-effector, traditional methods fail to address the redundancy that exists in an efficient way. The proposed method is singular, and in that it (1) Generates the most optimal angle vector in terms of maximum manipulability, a factor which determines the ease with which the arm moves, for a given end-vector. (2) Proposes a novel approach to inculcate the complexity of dealing with real coordinate system by proposing a machine learning technique that uses neural networks to predict angle vector for practically any end-effector position although it learns on only a few sampled space. (3) Works in real time since the entire optimization of the manipulability measure are done offline before training the neural network using a relevant technique which makes the proposed method suitable for practical uses. (4) It also determines the shortest, smooth path along which the manipulator can move along avoiding any obstacles. To the best of the authors’ knowledge, this is the first neural-net-based optimized inverse kinematics method applied for a robotic manipulator arm, and its optimal and simple structure also makes it possible to run it on NVIDIA Jetson Nano Module.