ArVoice comprises: (1) professionally recorded audio by 2 male and 2 female voice artists from diacritized transcripts, (2) professionally recorded audio by 1 male and 1 female voice artists from undiacritized transcripts, (3) a modified subset of the
33
22 commits
1 linked in READMEs
updated Oct 31, 2025
README
ArVoice: A Multi-Speaker Dataset for Arabic Speech Synthesis
Hawau Olamide Toyin, Rufael Marew, Humaid Alblooshi, Samar M. Magdy, Hanan Aldarmaki
{hawau.toyin, hanan.aldarmaki}@mbzuai.ac.ae
ArVoice is a multi-speaker Modern Standard Arabic (MSA) speech corpus with fully diacritized transcriptions, intended for multi-speaker speech synthesis, and can be useful for other tasks such as speech-based diacritic restoration, voice conversion, and deepfake detection.
ArVoice comprises: (1) professionally recorded audio by 2 male and 2 female voice artists from diacritized transcripts, (2) professionally recorded audio by 1 male and 1 female voice artists from undiacritized transcripts, (3) a modified subset of the
Arabic Speech Corpus, and (4) synthesized speech using commercial TTS systems. The complete corpus consists of a total of 83.52 hours of speech across 11 voices; around 10 hours consist of human voices from 7 speakers.
This repo consists of only Parts (3), ASC subset, and (4) synthetic subset ; to access the main subset, part (1,2), which consists of six professional speakers, please sign this agreement and email it to us.
If you use the dataset or transcriptions provided in Huggingface, place cite the paper.
Usage Example
df = load_dataset("MBZUAI/ArVoice", "Human_3") #data_dir options: Human_3, Synthetic,
print(df)
DatasetDict({
train: Dataset({
features: ['original_wav', 'normalized_wav', 'speaker_id', 'transcription'],
num_rows: 907
})
test: Dataset({
features: ['original_wav', 'normalized_wav', 'speaker_id', 'transcription'],
num_rows: 100
})
})
Data Statistics
| Type | Part | Gender | Speaker Origin | Duration (hrs) | Text Source |
|---|---|---|---|---|---|
| Human | ArVoice Part 1 | M | Egypt | 1.17 | Tashkeela |
| F | Jordan | 1.45 | |||
| M | Egypt | 1.58 | |||
| F | Morocco | 1.23 | |||
| ArVoice Part 2 | M | Palestine | 0.93 | Khaleej | |
| F | Egypt | 0.93 | |||
| ArVoice Part 3 | M | Syria | 2.69 | ASC | |
| Synthetic | ArVoice Part 4 | 2×M, 2×F | - | 73.5 | Tashkeela, Khaleej, ASC |
License: https://creativecommons.org/licenses/by/4.0/
Citation
@inproceedings{toyin25_interspeech,
title = {{ArVoice: A Multi-Speaker Dataset for Arabic Speech Synthesis}},
author = {Hawau Toyin and Rufael Marew and Humaid Alblooshi and Samar M. Magdy and Hanan Aldarmaki},
year = {2025},
booktitle = {{Interspeech 2025}},
pages = {4808--4812},
doi = {10.21437/Interspeech.2025-1550},
issn = {2958-1796},
}
Similar repos
Linked in READMEs
ArVoice comprises: (1) professionally recorded audio by 2 male and 2 female voice artists from diacritized transcripts, (2) professionally recorded audio by 1 male and 1 female voice artists from undiacritized transcripts, (3) a modified subset of the
33
22 commits
1 linked in READMEs
updated Oct 31, 2025
README
ArVoice: A Multi-Speaker Dataset for Arabic Speech Synthesis
Hawau Olamide Toyin, Rufael Marew, Humaid Alblooshi, Samar M. Magdy, Hanan Aldarmaki
{hawau.toyin, hanan.aldarmaki}@mbzuai.ac.ae
ArVoice is a multi-speaker Modern Standard Arabic (MSA) speech corpus with fully diacritized transcriptions, intended for multi-speaker speech synthesis, and can be useful for other tasks such as speech-based diacritic restoration, voice conversion, and deepfake detection.
ArVoice comprises: (1) professionally recorded audio by 2 male and 2 female voice artists from diacritized transcripts, (2) professionally recorded audio by 1 male and 1 female voice artists from undiacritized transcripts, (3) a modified subset of the
Arabic Speech Corpus, and (4) synthesized speech using commercial TTS systems. The complete corpus consists of a total of 83.52 hours of speech across 11 voices; around 10 hours consist of human voices from 7 speakers.
This repo consists of only Parts (3), ASC subset, and (4) synthetic subset ; to access the main subset, part (1,2), which consists of six professional speakers, please sign this agreement and email it to us.
If you use the dataset or transcriptions provided in Huggingface, place cite the paper.
Usage Example
df = load_dataset("MBZUAI/ArVoice", "Human_3") #data_dir options: Human_3, Synthetic,
print(df)
DatasetDict({
train: Dataset({
features: ['original_wav', 'normalized_wav', 'speaker_id', 'transcription'],
num_rows: 907
})
test: Dataset({
features: ['original_wav', 'normalized_wav', 'speaker_id', 'transcription'],
num_rows: 100
})
})
Data Statistics
| Type | Part | Gender | Speaker Origin | Duration (hrs) | Text Source |
|---|---|---|---|---|---|
| Human | ArVoice Part 1 | M | Egypt | 1.17 | Tashkeela |
| F | Jordan | 1.45 | |||
| M | Egypt | 1.58 | |||
| F | Morocco | 1.23 | |||
| ArVoice Part 2 | M | Palestine | 0.93 | Khaleej | |
| F | Egypt | 0.93 | |||
| ArVoice Part 3 | M | Syria | 2.69 | ASC | |
| Synthetic | ArVoice Part 4 | 2×M, 2×F | - | 73.5 | Tashkeela, Khaleej, ASC |
License: https://creativecommons.org/licenses/by/4.0/
Citation
@inproceedings{toyin25_interspeech,
title = {{ArVoice: A Multi-Speaker Dataset for Arabic Speech Synthesis}},
author = {Hawau Toyin and Rufael Marew and Humaid Alblooshi and Samar M. Magdy and Hanan Aldarmaki},
year = {2025},
booktitle = {{Interspeech 2025}},
pages = {4808--4812},
doi = {10.21437/Interspeech.2025-1550},
issn = {2958-1796},
}