FLA-Dataset: A Database of Four Location-Based Audio Commands in Brazilian Portuguese

Description FLA-Dataset is a wake word dataset in Brazilian Portuguese designed to train and evaluate keyword spotting and voice command recognition systems. It is suitable for applications such as virtual assistants, embedded systems, and robotics. Dataset Structure The dataset is organized into four main folders, one for each activation word: direita (right) esquerda (left) frente (forward) pare (stop) Each folder contains .wav audio files sampled at 16 kHz, with recordings of different speakers pronouncing the corresponding word. File Naming Convention Files follow the naming pattern: <word>_s<speaker_id>_a<sample_id>.wav sXXX indicates the speaker ID aYYY indicates the audio sample number for that speaker Speakers FLA-Dataset includes recordings from 30 speakers, with the following gender distribution: 10 female (IDs: s000 to s009) 20 male (IDs: s010 to s029) This speaker diversity supports more robust and generalizable models. Statistics The dataset contains a total of 2,329 audio samples, distributed as follows: direita: 624 samples esquerda: 514 samples frente: 597 samples pare: 594 samples Total duration: 40.78 minutesAverage duration per audio: 1.05 seconds Audio Specifications Format: WAV Sample rate: 16 kHz Type: Isolated words Language: Brazilian Portuguese Potential Applications Wake word detection Voice command recognition Embedded systems and voice-based interfaces in Portuguese