Arnaldo Candido Junior

Daniel Peixoto Pinto da Silva, Edresson Casanova, Lucas Rafael Stefanel Gris et al.

During the outbreak of COVID-19 pandemic, several research areas joined efforts to mitigate the damages caused by SARS-CoV-2. In this paper we present an interpretability analysis of a convolutional neural network based model for COVID-19 detection in audios. We investigate which features are important for model decision process, investigating spectrograms, F0, F0 standard deviation, sex and age. Following, we analyse model decisions by generating heat maps for the trained models to capture their attention during the decision process. Focusing on a explainable Inteligence Artificial approach, we show that studied models can taken unbiased decisions even in the presence of spurious data in the training set, given the adequate preprocessing steps. Our best model has 94.44% of accuracy in detection, with results indicating that models favors spectrograms for the decision process, particularly, high energy areas in the spectrogram related to prosodic domains, while F0 also leads to efficient COVID-19 detection.

Augusto Seben da Rosa, Marlon Daniel Angeli, Jorge Aikes Junior et al.

We developed a jitted compiler for training Artificial Neural Networks using C++, LLVM and Cuda. It features object-oriented characteristics, strong typing, parallel workers for data pre-processing, pythonic syntax for expressions, PyTorch like model declaration and Automatic Differentiation. We implement the mechanisms of cache and pooling in order to manage VRAM, cuBLAS for high performance matrix multiplication and cuDNN for convolutional layers. Our experiments with Residual Convolutional Neural Networks on ImageNet, we reach similar speed but degraded performance. Also, the GRU network experiments show similar accuracy, but our compiler have degraded speed in that task. However, our compiler demonstrates promising results at the CIFAR-10 benchmark, in which we reach the same performance and about the same speed as PyTorch. We make the code publicly available at: https://github.com/NoSavedDATA/NoSavedKaleidoscope

Augusto Seben da Rosa, Frederico Santos de Oliveira, Anderson da Silva Soares et al.

Computer vision in general presented several advances such as training optimizations, new architectures (pure attention, efficient block, vision language models, generative models, among others). This have improved performance in several tasks such as classification, and others. However, the majority of these models focus on modifications that are taking distance from realistic neuroscientific approaches related to the brain. In this work, we adopt a more bio-inspired approach and present the Yin Yang Convolutional Network, an architecture that extracts visual manifold, its blocks are intended to separate analysis of colors and forms at its initial layers, simulating occipital lobe's operations. Our results shows that our architecture provides State-of-the-Art efficiency among low parameter architectures in the dataset CIFAR-10. Our first model reached 93.32\% test accuracy, 0.8\% more than the older SOTA in this category, while having 150k less parameters (726k in total). Our second model uses 52k parameters, losing only 3.86\% test accuracy. We also performed an analysis on ImageNet, where we reached 66.49\% validation accuracy with 1.6M parameters. We make the code publicly available at: https://github.com/NoSavedDATA/YinYang_CNN.

Edresson Casanova, Christopher Shulby, Alexander Korolev et al.

We explore cross-lingual multi-speaker speech synthesis and cross-lingual voice conversion applied to data augmentation for automatic speech recognition (ASR) systems in low/medium-resource scenarios. Through extensive experiments, we show that our approach permits the application of speech synthesis and voice conversion to improve ASR systems using only one target-language speaker during model training. We also managed to close the gap between ASR models trained with synthesized versus human speech compared to other works that use many speakers. Finally, we show that it is possible to obtain promising ASR training results with our data augmentation method using only a single real speaker in a target language.

Marcelo Matheus Gauy, Natalia Hitomi Koza, Ricardo Mikio Morita et al.

We contrast high effectiveness of state of the art deep learning architectures designed for general audio classification tasks, refined for respiratory insufficiency (RI) detection and blood oxygen saturation (SpO$_2$) estimation and classification through automated audio analysis. Recently, multiple deep learning architectures have been proposed to detect RI in COVID patients through audio analysis, achieving accuracy above 95% and F1-score above 0.93. RI is a condition associated with low SpO$_2$ levels, commonly defined as the threshold SpO$_2$ <92%. While SpO$_2$ serves as a crucial determinant of RI, a medical doctor's diagnosis typically relies on multiple factors. These include respiratory frequency, heart rate, SpO$_2$ levels, among others. Here we study pretrained audio neural networks (CNN6, CNN10 and CNN14) and the Masked Autoencoder (Audio-MAE) for RI detection, where these models achieve near perfect accuracy, surpassing previous results. Yet, for the regression task of estimating SpO$_2$ levels, the models achieve root mean square error values exceeding the accepted clinical range of 3.5% for finger oximeters. Additionally, Pearson correlation coefficients fail to surpass 0.3. As deep learning models perform better in classification than regression, we transform SpO$_2$-regression into a SpO$_2$-threshold binary classification problem, with a threshold of 92%. However, this task still yields an F1-score below 0.65. Thus, audio analysis offers valuable insights into a patient's RI status, but does not provide accurate information about actual SpO$_2$ levels, indicating a separation of domains in which voice and speech biomarkers may and may not be useful in medical diagnostics under current technologies.

Lucas Rafael Stefanel Gris, Arnaldo Candido Junior, Vinícius G. dos Santos et al.

The NURC Project that started in 1969 to study the cultured linguistic urban norm spoken in five Brazilian capitals, was responsible for compiling a large corpus for each capital. The digitized NURC/SP comprises 375 inquiries in 334 hours of recordings taken in São Paulo capital. Although 47 inquiries have transcripts, there was no alignment between the audio-transcription, and 328 inquiries were not transcribed. This article presents an evaluation and error analysis of three automatic speech recognition models trained with spontaneous speech in Portuguese and one model trained with prepared speech. The evaluation allowed us to choose the best model, using WER and CER metrics, in a manually aligned sample of NURC/SP, to automatically transcribe 284 hours.

Julio Cesar Galdino, Sidney Evaldo Leal, Leticia Gabriella De Souza et al.

Spontaneous speech presents several challenges for speech synthesis, particularly in capturing the natural flow of conversation, including turn-taking, pauses, and disfluencies. Although speech synthesis systems have made significant progress in generating natural and intelligible speech, primarily through architectures that implicitly model prosodic features such as pitch, intensity, and duration, the construction of datasets with explicit prosodic segmentation and their impact on spontaneous speech synthesis remains largely unexplored. This paper evaluates the effects of manual and automatic prosodic segmentation annotations in Brazilian Portuguese on the quality of speech synthesized by a non-autoregressive model, FastSpeech 2. Experimental results show that training with prosodic segmentation produced slightly more intelligible and acoustically natural speech. While automatic segmentation tends to create more regular segments, manual prosodic segmentation introduces greater variability, which contributes to more natural prosody. Analysis of neutral declarative utterances showed that both training approaches reproduced the expected nuclear accent pattern, but the prosodic model aligned more closely with natural pre-nuclear contours. To support reproducibility and future research, all datasets, source codes, and trained models are publicly available under the CC BY-NC-ND 4.0 license.

Rodrigo Lima, Sidney Evaldo Leal, Arnaldo Candido Junior et al.

We present a freely available spontaneous speech corpus for the Brazilian Portuguese language and report preliminary automatic speech recognition (ASR) results, using both the Wav2Vec2-XLSR-53 and Distil-Whisper models fine-tuned and trained on our corpus. The NURC-SP Audio Corpus comprises 401 different speakers (204 females, 197 males) with a total of 239.30 hours of transcribed audio recordings. To the best of our knowledge, this is the first large Paulistano accented spontaneous speech corpus dedicated to the ASR task in Portuguese. We first present the design and development procedures of the NURC-SP Audio Corpus, and then describe four ASR experiments in detail. The experiments demonstrated promising results for the applicability of the corpus for ASR. Specifically, we fine-tuned two versions of Wav2Vec2-XLSR-53 model, trained a Distil-Whisper model using our dataset with labels determined by Whisper Large-V3 model, and fine-tuned this Distil-Whisper model with our corpus. Our best results were the Distil-Whisper fine-tuned over NURC-SP Audio Corpus with a WER of 24.22% followed by a fine-tuned versions of Wav2Vec2-XLSR-53 model with a WER of 33.73%, that is almost 10% point worse than Distil-Whisper's. To enable experiment reproducibility, we share the NURC-SP Audio Corpus dataset, pre-trained models, and training recipes in Hugging-Face and Github repositories.

Arnaldo Candido Junior, Edresson Casanova, Anderson Soares et al.

Automatic Speech recognition (ASR) is a complex and challenging task. In recent years, there have been significant advances in the area. In particular, for the Brazilian Portuguese (BP) language, there were about 376 hours public available for ASR task until the second half of 2020. With the release of new datasets in early 2021, this number increased to 574 hours. The existing resources, however, are composed of audios containing only read and prepared speech. There is a lack of datasets including spontaneous speech, which are essential in different ASR applications. This paper presents CORAA (Corpus of Annotated Audios) v1. with 290.77 hours, a publicly available dataset for ASR in BP containing validated pairs (audio-transcription). CORAA also contains European Portuguese audios (4.69 hours). We also present a public ASR model based on Wav2Vec 2.0 XLSR-53 and fine-tuned over CORAA. Our model achieved a Word Error Rate of 24.18% on CORAA test set and 20.08% on Common Voice test set. When measuring the Character Error Rate, we obtained 11.02% and 6.34% for CORAA and Common Voice, respectively. CORAA corpora were assembled to both improve ASR models in BP with phenomena from spontaneous speech and motivate young researchers to start their studies on ASR for Portuguese. All the corpora are publicly available at https://github.com/nilc-nlp/CORAA under the CC BY-NC-ND 4.0 license.

Daniel Oliveira de Brito, Letícia Gabriella de Souza, Marcelo Matheus Gauy et al.

This technical report investigates the performance of pre-trained audio models on COVID-19 detection tasks using established benchmark datasets. We fine-tuned Audio-MAE and three PANN architectures (CNN6, CNN10, CNN14) on the Coswara and COUGHVID datasets, evaluating both intra-dataset and cross-dataset generalization. We implemented a strict demographic stratification by age and gender to prevent models from exploiting spurious correlations between demographic characteristics and COVID-19 status. Intra-dataset results showed moderate performance, with Audio-MAE achieving the strongest result on Coswara (0.82 AUC, 0.76 F1-score), while all models demonstrated limited performance on Coughvid (AUC 0.58-0.63). Cross-dataset evaluation revealed severe generalization failure across all models (AUC 0.43-0.68), with Audio-MAE showing strong performance degradation (F1-score 0.00-0.08). Our experiments demonstrate that demographic balancing, while reducing apparent model performance, provides more realistic assessment of COVID-19 detection capabilities by eliminating demographic leakage - a confounding factor that inflate performance metrics. Additionally, the limited dataset sizes after balancing (1,219-2,160 samples) proved insufficient for deep learning models that typically require substantially larger training sets. These findings highlight fundamental challenges in developing generalizable audio-based COVID-19 detection systems and underscore the importance of rigorous demographic controls for clinically robust model evaluation.

Lucas Rafael Stefanel Gris, Ricardo Marcacini, Arnaldo Candido Junior et al.

Automatic speech recognition (ASR) systems play a key role in applications involving human-machine interactions. Despite their importance, ASR models for the Portuguese language proposed in the last decade have limitations in relation to the correct identification of punctuation marks in automatic transcriptions, which hinder the use of transcriptions by other systems, models, and even by humans. However, recently Whisper ASR was proposed by OpenAI, a general-purpose speech recognition model that has generated great expectations in dealing with such limitations. This chapter presents the first study on the performance of Whisper for punctuation prediction in the Portuguese language. We present an experimental evaluation considering both theoretical aspects involving pausing points (comma) and complete ideas (exclamation, question, and fullstop), as well as practical aspects involving transcript-based topic modeling - an application dependent on punctuation marks for promising performance. We analyzed experimental results from videos of Museum of the Person, a virtual museum that aims to tell and preserve people's life histories, thus discussing the pros and cons of Whisper in a real-world scenario. Although our experiments indicate that Whisper achieves state-of-the-art results, we conclude that some punctuation marks require improvements, such as exclamation, semicolon and colon.

Edresson Casanova, Julian Weber, Christopher Shulby et al.

YourTTS brings the power of a multilingual approach to the task of zero-shot multi-speaker TTS. Our method builds upon the VITS model and adds several novel modifications for zero-shot multi-speaker and multilingual training. We achieved state-of-the-art (SOTA) results in zero-shot multi-speaker TTS and results comparable to SOTA in zero-shot voice conversion on the VCTK dataset. Additionally, our approach achieves promising results in a target language with a single-speaker dataset, opening possibilities for zero-shot multi-speaker TTS and zero-shot voice conversion systems in low-resource languages. Finally, it is possible to fine-tune the YourTTS model with less than 1 minute of speech and achieve state-of-the-art results in voice similarity and with reasonable quality. This is important to allow synthesis for speakers with a very different voice or recording characteristics from those seen during training.

Lucas Rafael Stefanel Gris, Edresson Casanova, Frederico Santos de Oliveira et al.

Deep learning techniques have been shown to be efficient in various tasks, especially in the development of speech recognition systems, that is, systems that aim to transcribe an audio sentence in a sequence of written words. Despite the progress in the area, speech recognition can still be considered difficult, especially for languages lacking available data, such as Brazilian Portuguese (BP). In this sense, this work presents the development of an public Automatic Speech Recognition (ASR) system using only open available audio data, from the fine-tuning of the Wav2vec 2.0 XLSR-53 model pre-trained in many languages, over BP data. The final model presents an average word error rate of 12.4% over 7 different datasets (10.5% when applying a language model). According to our knowledge, the obtained error is the lowest among open end-to-end (E2E) ASR models for BP.

Edresson Casanova, Christopher Shulby, Eren Gölge et al.

In this paper, we propose SC-GlowTTS: an efficient zero-shot multi-speaker text-to-speech model that improves similarity for speakers unseen during training. We propose a speaker-conditional architecture that explores a flow-based decoder that works in a zero-shot scenario. As text encoders, we explore a dilated residual convolutional-based encoder, gated convolutional-based encoder, and transformer-based encoder. Additionally, we have shown that adjusting a GAN-based vocoder for the spectrograms predicted by the TTS model on the training dataset can significantly improve the similarity and speech quality for new speakers. Our model converges using only 11 speakers, reaching state-of-the-art results for similarity with new speakers, as well as high speech quality.

Edresson Casanova, Arnaldo Candido Junior, Christopher Shulby et al.

Speech provides a natural way for human-computer interaction. In particular, speech synthesis systems are popular in different applications, such as personal assistants, GPS applications, screen readers and accessibility tools. However, not all languages are on the same level when in terms of resources and systems for speech synthesis. This work consists of creating publicly available resources for Brazilian Portuguese in the form of a novel dataset along with deep learning models for end-to-end speech synthesis. Such dataset has 10.5 hours from a single speaker, from which a Tacotron 2 model with the RTISI-LA vocoder presented the best performance, achieving a 4.03 MOS value. The obtained results are comparable to related works covering English language and the state-of-the-art in Portuguese.

Edresson Casanova, Arnaldo Candido Junior, Christopher Shulby et al.

In this paper we present an efficient method for training models for speaker recognition using small or under-resourced datasets. This method requires less data than other SOTA (State-Of-The-Art) methods, e.g. the Angular Prototypical and GE2E loss functions, while achieving similar results to those methods. This is done using the knowledge of the reconstruction of a phoneme in the speaker's voice. For this purpose, a new dataset was built, composed of 40 male speakers, who read sentences in Portuguese, totaling approximately 3h. We compare the three best architectures trained using our method to select the best one, which is the one with a shallow architecture. Then, we compared this model with the SOTA method for the speaker recognition task: the Fast ResNet-34 trained with approximately 2,000 hours, using the loss functions Angular Prototypical and GE2E. Three experiments were carried out with datasets in different languages. Among these three experiments, our model achieved the second best result in two experiments and the best result in one of them. This highlights the importance of our method, which proved to be a great competitor to SOTA speaker recognition models, with 500x less data and a simpler approach.