Recently, several public datasets for automatic speech recognition (ASR) in Brazilian Portuguese (BP) have been released, improving ASR systems performance. However, these datasets lack diversity in terms of age groups, regional accents, and education levels. In this paper, we present a new publicly available dataset consisting of 289 life story interviews (365 hours), featuring a broad range of speakers varying in age, education, and regional accents. First, we demonstrated the presence of bias in current BP ASR models concerning education levels and age groups. Second, we showed that our dataset helps mitigate these biases. Additionally, an ASR model trained on our dataset performed better during evaluation on a diverse test set. Finally, the ASR model trained with our dataset was extrinsically evaluated through a topic modeling task that utilized the automatically transcribed output.

Autoregressive speech token generation models produce speech with remarkable variety and naturalness but often suffer from hallucinations and undesired vocalizations that do not conform to conditioning inputs. To address these challenges, we introduce Koel-TTS, an encoder-decoder transformer model for multilingual TTS that improves contextual adherence of speech generation LLMs through preference alignment and classifier-free guidance (CFG). For preference alignment, we design a reward system that ranks model outputs using automatic metrics derived from speech recognition and speaker verification models, encouraging generations that better match the input text and speaker identity. CFG further allows fine-grained control over the influence of conditioning inputs during inference by interpolating conditional and unconditional logits. Notably, applying CFG to a preference-aligned model yields additional gains in transcription accuracy and speaker similarity, demonstrating the complementary benefits of both techniques. Koel-TTS achieves state-of-the-art results in zero-shot TTS, outperforming prior LLM-based models on intelligibility, speaker similarity, and naturalness, despite being trained on significantly less data.

Automatic analysis of connected speech by natural language processing techniques is a promising direction for diagnosing cognitive impairments. However, some difficulties still remain: the time required for manual narrative transcription and the decision on how transcripts should be divided into sentences for successful application of parsers used in metrics, such as Idea Density, to analyze the transcripts. The main goal of this paper was to develop a generic segmentation system for narratives of neuropsychological language tests. We explored the performance of our previous single-dataset-trained sentence segmentation architecture in a richer scenario involving three new datasets used to diagnose cognitive impairments, comprising different stories and two types of stimulus presentation for eliciting narratives — visual and oral — via illustrated story-book and sequence of scenes, and by retelling. Also, we proposed and evaluated three modifications to our previous RCNN architecture: (i) the inclusion of a Linear Chain CRF; (ii) the inclusion of a self-attention mechanism; and (iii) the replacement of the LSTM recurrent layer by a Quasi-Recurrent Neural Network layer. Our study allowed us to develop two new models for segmenting impaired speech transcriptions, along with an ideal combination of datasets and specific groups of narratives to be used as the training set.