voa-engines/dataset_default_curator_1ksp

VOA PT-BR Whisper Dataset

Dataset Description

This dataset contains audio segments processed with Whisper ASR model for the Portuguese language. The Default VAD configuration was used to generate speech segments.

Each segment has an audio file (.wav) and associated metadata including:

• Transcription (generated using Whisper)
• Duration
• Start and end times
• Audio ID and segment ID

Usage

This dataset is intended for:

1. Evaluation of ASR models for Portuguese
2. Fine-tuning ASR models for Portuguese
3. Benchmark for speech recognition tasks

Dataset Structure

• audio/: Contains WAV files of audio segments
• whisper_dataset.csv: Original dataset with metadata
• whisper_dataset_with_paths.csv: Dataset with updated audio paths
• metadata.json: Metadata in JSON format for HuggingFace
• dataset_dict.json: Dataset in HuggingFace compatible format

Loading the Dataset

You can load this dataset using the HuggingFace Datasets library:

from datasets import load_dataset

# Load from the Hub
dataset = load_dataset("voa-engines/whisper-default-dataset_1ksp")

# Access the first example
example = dataset["train"][0]
print(f"Audio file: {example['audio']}")
print(f"Transcription: {example['transcription']}")

Created on: 2025-04-30