naijavoices/naijavoices-dataset-compressed

Introduction

Welcome to the NaijaVoices dataset. The NaijaVoices dataset consisted of 1,500 hours of authentic speech (from over 5,000 diverse speakers!) and expert curated text in Igbo, Hausa, and Yoruba. 500 hours for each of the three languages. It also boasts of adequate female representation and balanced age-range distribution (young to old speakers). For more about the dataset info visit our website: https://naijavoices.com/

Metadata

The metadata.csv file contains the metadata for all the audio files. Each row contains:

audio: the audio filename
text: the transcript of the audio
speaker_id: the id of the speaker. This proves useful for ethical TTS use cases
language: the language of the transcript (whether Igbo or Hausa or Yoruba)
age_bracket: the age-range  of the speaker
gender: the gender of the speaker

Load the csv like this:

import pandas as pd
df = pd.read_csv('metadata.csv',sep='~')

Usage

You must read and accept our data usage guidelines before using the NaijaVoices dataset. If you encounter any issues with the processes below, please create a discussion here and we will help.

1: Unzipping the audio files

The audio files are zipped in split zip files (audio-files.zip and audio-files.z01) due to storage limits. Therefore, follow the process below to correctly unzip and access the audio files:

1. First of all, you need the zip package. You should have this already installed with many Linux distributions. If you need to install it, refer to here for a comprehensive guide on installing it.
2. Since the zipping was split, we need to first use the zip command to combine the split zip files into a single zip archive (let's call it audio.zip). You do that by running zip -F audio-files.zip --out audio.zip
3. Finally use unzip package to unzip the audio.zip file: unzip audio.zip. Voilà! You can now view the audio files.

2: Loading the audio files

To load the audio files in python, we have found the following process to work best given the nature of the audio data:

1. Use librosa package. You can install librosa with pip install librosa.
2. You also need to install ffmpeg because librosa will use the codec of ffmpeg. On Linux, you can simply do sudo apt install ffmpeg. You can refer here for a guide on installing ffmpeg on Windows OS.
3. Finally, to load an audio simply do:

    audio_path = os.path.join(AUDIO_FOLDER,AUDIO_FILENAME) # AUDIO_FILENAME is from the `audio` column of the metadata.csv and AUDIO_FOLDER is the path to the unzipped audio file.
    audio, sr = librosa.load(audio_path, sr=48_000)

License

Our dataset is licensed under the CC BY-NC-SA 4.0 license. Basically, It's free for personal and research use, with proper credit to the community. For commercial interests, please reach out to us.