mikelalda/common_voice_17_0_eu

tags:

• —audio
• —automatic-speech-recognition
• —common-voice
• —euskera
• —eu license: cc0-1.0 ---

Common Voice 17.0 - Euskera with Audio Features

This dataset contains the Euskera (Basque) subset of the Mozilla Common Voice 17.0 dataset. It includes audio clips and corresponding transcriptions, prepared with audio features readily available for use with libraries like Hugging Face datasets. For the preparation of this dataset, the following notebook was used: Preparar_dataset_euskera.ipynb

Source: The original data is from the Mozilla Common Voice 17.0 dataset.

Language: Euskera (eu)

Data Format: The dataset is provided in Parquet format. The main features include:

• —client_id: A unique identifier for the speaker.
• —path: The original filename of the audio clip.
• —sentence_id: A unique identifier for the sentence.
• —sentence: The transcription of the audio clip.
• —up_votes, down_votes: Community validation votes.
• —age, gender, accents, variant: Speaker metadata (may contain missing values).
• —locale: Language locale code ('eu').
• —split: Original data split segment (e.g., 'train', 'test', 'dev', 'validated', 'invalidated', 'other').
• —audio_file: The audio file (this column is cast as an Audio feature when loaded with datasets).

Splits: This dataset contains a 'train' split, which combines the original 'train', 'test', 'dev', 'validated', 'invalidated', and 'other' splits from Common Voice 17.0 for Euskera. The original 'segment' column is retained to allow for splitting if needed.

Data Collection: The data was collected as part of the Mozilla Common Voice project, a crowdsourcing initiative to build a free and open dataset of recorded speech.

Licensing: This dataset is licensed under the Creative Commons CC0 1.0 Universal Public Domain Dedication, the same as the original Common Voice dataset.

Intended Use: This dataset is intended for research and development in speech technologies, particularly for Euskera Automatic Speech Recognition (ASR) and other audio-related tasks.

Known Limitations:

• —The dataset may contain noise, variations in audio quality, and inaccuracies in transcriptions inherent in crowdsourced data.
• —Speaker metadata (age, gender, accents) may be incomplete or inaccurate.
• —The dataset size for Euskera may be limited compared to more widely spoken languages.

For more details on the original Common Voice dataset, please refer to its documentation on the Hugging Face Hub.