Hi! I'm Bernardo. I work as a machine learning researcher for applications on audio and music.
I recently completed my PhD at Telecom Paris, supervised by prof. Gaël Richard and prof. Geoffroy Peeters, as part of the ADASP group and the HI-Audio project. My thesis was on audio source separation and music information retrieval, with a particular interest in self-supervised and unsupervised methods for training smart and efficient models.
Along the way, I was a research scientist intern at Deezer Research and Sony CSL (Music Team). I hold a Master's in Artificial Intelligence from the MVA program at ENS Paris-Saclay, an engineering degree from Telecom Paris, and a Bachelor's in Electrical and Computer Engineering from Universidade Federal de Minas Gerais.
News
- Jul 2026 Defended my PhD at Telecom Paris! 🎉 The manuscript, Structure-Guided Learning for Audio Analysis under Low Supervision, is available online.
Selected publications
-
ICASSP 2026 Learning Linearity in Audio Consistency Autoencoders via Implicit Regularization
Audio autoencoders learn useful, compressed audio representations, but their non-linear latent spaces prevent intuitive algebraic manipulation such as mixing or scaling. We introduce a simple training methodology to induce linearity in a high-compression Consistency Autoencoder (CAE) by using data augmentation, thereby inducing homogeneity (equivariance to scalar gain) and additivity (the decoder preserves addition) without altering the model's architecture or loss function. When trained with our method, the CAE exhibits linear behavior in both the encoder and decoder while preserving reconstruction fidelity. We test the practical utility of our learned space on music source composition and separation via simple latent arithmetic. This work presents a straightforward technique for constructing structured latent spaces, enabling more intuitive and efficient audio processing. -
TASLP 2025 The Inverse Drum Machine: Source Separation Through Joint Transcription and Analysis-by-Synthesis
We present the Inverse Drum Machine (IDM), a novel approach to drum source separation that leverages an analysis-by-synthesis framework combined with deep learning. Unlike recent supervised methods that require isolated stem recordings, our approach operates on drum mixtures with only transcription annotations. IDM integrates Automatic Drum Transcription and One-shot Drum Sample Synthesis, jointly optimizing these tasks in an end-to-end manner. By convolving synthesized one-shot samples with estimated onsets—akin to a drum machine—we reconstruct the individual drum stems and train a deep neural network on the reconstruction of the mixture. Experiments on the StemGMD dataset demonstrate that IDM achieves separation quality comparable to state-of-the-art supervised methods that require isolated stems, while significantly outperforming matrix decomposition baselines. -
TISMIR 2025 PESTO: Real-Time Pitch Estimation with Self-supervised Transposition-equivariant Objective
In this paper, we introduce PESTO, a self-supervised learning approach for single-pitch estimation using a Siamese architecture. Our model processes individual frames of a Variable-Q Transform (VQT) and predicts pitch distributions. The neural network is designed to be equivariant to translations, notably thanks to a Toeplitz fully-connected layer. In addition, we construct pitch-shifted pairs by translating and cropping the VQT frames and train our model with a novel class-based transposition-equivariant objective, eliminating the need for annotated data. Thanks to this architecture and training objective, our model achieves remarkable performances while being very lightweight (130k parameters). Evaluations on music and speech datasets (MIR-1K, MDB-stem-synth, and PTDB) demonstrate that PESTO not only outperforms self-supervised baselines but also competes with supervised methods, exhibiting superior cross-dataset generalization. Finally, we enhance PESTO's practical utility by developing a streamable VQT implementation using cached convolutions. Combined with our model's low latency (less than 10 ms) and minimal parameter count, this makes PESTO particularly suitable for real-time applications. -
ICASSP 2024 Unsupervised Harmonic Parameter Estimation Using Differentiable DSP and Spectral Optimal Transport
In neural audio signal processing, pitch conditioning has been used to enhance the performance of synthesizers. However, jointly training pitch estimators and synthesizers is a challenge when using standard audio-to-audio reconstruction loss, leading to reliance on external pitch trackers. To address this issue, we propose using a spectral loss function inspired by optimal transportation theory that minimizes the displacement of spectral energy. We validate this approach through an unsupervised autoencoding task that fits a harmonic template to harmonic signals. We jointly estimate the fundamental frequency and amplitudes of harmonics using a lightweight encoder and reconstruct the signals using a differentiable harmonic synthesizer. The proposed approach offers a promising direction for improving unsupervised parameter estimation in neural audio applications. -
ISMIR 2023 Singer Identity Representation Learning using Self-Supervised Techniques
Significant strides have been made in creating voice identity representations using speech data. However, the same level of progress has not been achieved for singing voices. To bridge this gap, we suggest a framework for training singer identity encoders to extract representations suitable for various singing-related tasks, such as singing voice similarity and synthesis. We explore different selfsupervised learning techniques on a large collection of isolated vocal tracks and apply data augmentations during training to ensure that the representations are invariant to pitch and content variations. We evaluate the quality of the resulting representations on singer similarity and identification tasks across multiple datasets, with a particular emphasis on out-of-domain generalization. Our proposed framework produces high-quality embeddings that outperform both speaker verification and wav2vec 2.0 pre-trained baselines on singing voice while operating at 44.1 kHz. We release our code and trained models to facilitate further research on singing voice and related areas.
Full list on Google Scholar →