Company : AGM Telecom
Project 1: SAPDA (Sistema de Acessibilidade para Pessoa com Deficiência Auditiva)
-
This project was set to develop a ASR system to be inserted over telephonic services, with the aim to help people with hearing problems about using these services. The objective was to create a product that converts speech into text messages, to afford the communication between people without hearing problems and the ones who suffers it. It was a 2 years project and it was granted by FINEP.
Project 2: CORPORA AGM
-
This project was created to complement SAPDA project. The purpose was to build a acoustic and textual corpora to feed the models that were used on SAPDA's ASR. There was a team of annotators which were in charge to transcribe audio into text, collect textual material over the internet and evaluate the quality of these material. All the annotators were graduated and pos-gradutated students from UFG and UFSC, and there were granted by a RHAE/CNPQ project. It was a 2 years project.
Felipe's Role:
-
Work with linguistic computational models to enable computers to recognize spoken language ( ASR).
-
Mining, collecting, analyzing and processing a huge amount of Brazilian Portuguese speech and text data.
-
Defining the phonetic settings to a BP's ASR system
-
Corpora AGM coordinator
-
Selection of candidates to annotators role
-
Training the selected annotators
-
Definition of week activities
-
Evaluation about the quality of the generated material
-
-
Managing tools like MITLIN, HTK toolkit, Praat, Transcriber, Audacity
-
Creating shell scripts for text and audio manipulation
-
Developing text database for audio recording focused on ASR corpora (NLP)
-
Speech database recording
-
Grammar rules for NLP on Portuguese
-
Creating linguistic label for acoustic and textual data to cover linguistic issues like hesitation, filled pauses and spontaneous speech patterns
Link: sapda.appointy.com