The FalaBrasil Group


Creating and sharing free resources for speech and natural language processing in Brazilian Portuguese

FalaBrasil is a research group from Federal University of Pará (UFPA) in Belém, Brazil. The group has traditionally followed an open policy of creating and sharing non-profitable natural language tools and resources for speech recognition and speech synthesis in Brazilian Portuguese.

The idea concerning the FalaBrasil project started around 2000 with Professor Dr. Aldebaro Klautau (a PhD student at the time), who was responsible for the group’s foundation within the Signal Processing Laboratory (LaPS). Prof. Aldebaro is now director of the Telecommunications, Automation and Electronics Research and Development Center (LASSE). Ever since 2009, Klautau’s then-PhD-student Professor Dr. Nelson Neto has led the FalaBrasil group, which is currently part of the Laboratory of Visualization, Interaction and Intelligent Systems (LabVIS).

Regarding automatic speech recognition (ASR), the group has spent a lot of effort in developing resources for some of the most-known open-source platforms such as HTK, Julius, CMU Sphinx and, more recently, Kaldi. On the other hand, text-to-speech (TTS), or speech synthesis, is currently being implemented over the HTS toolkit. Other works involving speech generation techniques have been undergoing using the Klatt formant-based synthesizer.

Advances in natural language processing (NLP) have also been made. To date, the group has a rule-based grapheme-to-phoneme (G2P) converter, a grapheme and phoneme syllabification algorithm, a stress vowel identifier and a heteronym disambiguator software. Phonetic aligners with HDecode and Kaldi decoders have also been developed.

TODO: apps: coruja, fft-transcriber, simon, ...