Abstract:We recently presented a new model for singing synthesis based on a modified version of the WaveNet architecture. Instead of modeling raw waveform, we model features produced by a parametric vocoder that separates the influence of pitch and timbre. This allows conveniently modifying pitch to match any target melody, facilitates training on more modest dataset sizes, and significantly reduces training and generation times. Nonetheless, compared to modeling waveform directly, ways of effectively handling higher-dimensional outputs, multiple feature streams and regularization become more important with our approach. In this work, we extend our proposed system to include additional components for predicting F0 and phonetic timings from a musical score with lyrics. These expression-related features are learned together with timbrical features from a single set of natural songs. We compare our method to existing statistical parametric, concatenative, and neural network-based approaches using quantitative metrics as well as listening tests.

No abstract

We propose a sequence-to-sequence singing synthesizer, which avoids the need for training data with pre-aligned phonetic and acoustic features. Rather than the more common approach of a content-based attention mechanism combined with an autoregressive decoder, we use a different mechanism suitable for feed-forward synthesis. Given that phonetic timings in singing are highly constrained by the musical score, we derive an approximate initial alignment with the help of a simple duration model. Then, using a decoder based on a feed-forward variant of the Transformer model, a series of self-attention and convolutional layers refines the result of the initial alignment to reach the target acoustic features. Advantages of this approach include faster inference and avoiding the exposure bias issues that affect autoregressive models trained by teacher forcing. We evaluate the effectiveness of this model compared to an autoregressive baseline, the importance of self-attention, and the importance of the accuracy of the duration model.

We present a deep neural network based singing voice synthesizer, inspired by the Deep Convolutions Generative Adversarial Networks (DCGAN) architecture and optimized using the Wasserstein-GAN algorithm. We use vocoder parameters for acoustic modelling, to separate the influence of pitch and timbre. This facilitates the modelling of the large variability of pitch in the singing voice. Our network takes a block of consecutive frame-wise linguistic and fundamental frequency features, along with global singer identity as input and outputs vocoder features, corresponding to the block of features. This block-wise approach, along with the training methodology allows us to model temporal dependencies within the features of the input block. For inference, sequential blocks are concatenated using an overlap-add procedure. We show that the performance of our model is competitive with regards to the state-of-the-art and the original sample using objective metrics and a subjective listening test. We also present examples of the synthesis on a supplementary website and the source code via GitHub.

We present a new model for singing synthesis based on a modified version of the WaveNet architecture. Instead of modeling raw waveform, we model features produced by a parametric vocoder that separates the influence of pitch and timbre. This allows conveniently modifying pitch to match any target melody, facilitates training on more modest dataset sizes, and significantly reduces training and generation times. Our model makes frame-wise predictions using mixture density outputs rather than categorical outputs in order to reduce the required parameter count. As we found overfitting to be an issue with the relatively small datasets used in our experiments, we propose a method to regularize the model and make the autoregressive generation process more robust to prediction errors. Using a simple multi-stream architecture, harmonic, aperiodic and voiced/unvoiced components can all be predicted in a coherent manner. We compare our method to existing parametric statistical and state-of-the-art concatenative methods using quantitative metrics and a listening test. While naive implementations of the autoregressive generation algorithm tend to be inefficient, using a smart algorithm we can greatly speed up the process and obtain a system that's competitive in both speed and quality.

This article deals with automatic transcription of flamenco music recordings-more specifically, a cappella singing. We first study the specifics of flamenco singing and propose a transcription system based on fundamental frequency and energy estimation, which incorporates an iterative strategy for note segmentation and labeling. The proposed approach is evaluated on a music collection of 72 performances, including a variety of singers and recording conditions, and the presence or absence of percussion, background voices, and noise. We obtain satisfying results for the different approaches tested, and our system outperforms a state-of-the-art approach designed for other singing styles. In this study, we discuss the difficulties found in transcribing flamenco singing and in evaluating the obtained transcriptions, we analyze the influence of the different steps of the algorithm, and we state the main limitations of our approach and discuss challenges for future studies.

Latent generative models can learn higher-level underlying factors from complex data in an unsupervised manner. Such models can be used in a wide range of speech processing applications, including synthesis, transformation and classification. While there have been many advances in this field in recent years, the application of the resulting models to speech processing tasks is generally not explicitly considered. In this paper we apply the variational autoencoder (VAE) to the task of modeling frame-wise spectral envelopes. The VAE model has many attractive properties such as continuous latent variables, prior probability over these latent variables, a tractable lower bound on the marginal log likelihood, both generative and recognition models, and end-to-end training of deep models. We consider different aspects of training such models for speech data and compare them to more conventional models such as the Restricted Boltzmann Machine (RBM). While evaluating generative models is difficult, we try to obtain a balanced picture by considering both performance in terms of reconstruction error and when applying the model to a series of modeling and transformation tasks to get an idea of the quality of the learned features.

Sample and statistically based singing synthesizers typically require a large amount of data for automatically generating expressive synthetic performances. In this paper we present a singing synthesizer that using two rather small databases is able to generate expressive synthesis from an input consisting of notes and lyrics. The system is based on unit selection and uses the Wide-Band Harmonic Sinusoidal Model for transforming samples. The first database focuses on expression and consists of less than 2 minutes of free expressive singing using solely vowels. The second one is the timbre database which for the English case consists of roughly 35 minutes of monotonic singing of a set of sentences, one syllable per beat. The synthesis is divided in two steps. First, an expressive vowel singing performance of the target song is generated using the expression database. Next, this performance is used as input control of the synthesis using the timbre database and the target lyrics. A selection of synthetic performances have been submitted to the Interspeech Singing Synthesis Challenge 2016, in which they are compared to other competing systems.