BRAIN: A speech to text transcription approach based on Romanian corpus

Speech recognition applications enable the recognition and translation of spoken languages into text by computers. Due to the fact that the traditional approaches are costly and time consuming, the worldwide industry adopted speech recognition systems.

Andrei Scutelnicu, Anca Diana Bibiri, Mihaela Onofrei, and Mircea Hulea will present in their paper “A speech to text transcription approach based on Romanian corpus” a simple and efficient method for speech to text recognition, based on a machine learning approach, using a Romanian speech corpus.

Main components of a speech recognition system

The researchers used information from two questionnaires (AMPER-ROM[ANIA] and AMPRom) that belong to the project AMPRom (Romanian multimedia prosodic atlas), highlighting that this is the first prosodic atlas whose main purpose is to present the intonation patterns of the Romanian language.

Following the implementation of their algorithm, they conducted several tests and placed the recognition rate between 62% and 68%. Their next step is to add more features and test the method on a greater variety of data.

The upcoming issue of BRAIN Journal will provide more details on how their speech recognition system works.

Emilia Șteț