Development of Hausa dataset a baseline for speech recognition

Area of Science:

• Computational Linguistics
• Speech Technology
• African Languages

Background:

• The development of robust speech technologies requires extensive, high-quality datasets.
• Limited resources currently exist for low-resource languages like Hausa.
• Native speaker recordings are crucial for accurate speech model training.

Purpose of the Study:

• To create a comprehensive read-speech dataset for the Hausa language.
• To facilitate research and development in automatic speech recognition (ASR) and text-to-speech (TTS) systems for Hausa.
• To provide a segmented dataset (unigram and bigram) for advanced linguistic analysis.

Main Methods:

• Recruited native Hausa speakers for audio recording.
• Utilized professional audio studios at Nile University of Nigeria.
• Segmented recorded audio into unigram and bigram units.
• Collected a total of 47 hours of audio data.

Main Results:

• Successfully created a 47-hour Hausa read-speech dataset.
• The dataset is segmented into unigram and bigram units.
• The data was collected in controlled studio environments.

Conclusions:

• The Hausa speech dataset is a valuable resource for the research community.
• This dataset will accelerate the development of speech technologies for Hausa.
• It enables applications such as speech-to-text and text-to-speech for the Hausa language.