You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

I acknowledge that this dataset contains synthetic data generated through the process described in this paper and its corresponding GitHub codebase. It is not generated by humans, and as such, may contain model-induced biases.
The synthetic text data in this dataset was generated using Claude Sonnet 3.7. I agree to use the content only for non-commercial purposes, and in accordance with Anthropic’s Terms of Service and Usage Policy. This includes not using the outputs to develop products or services that compete with Anthropic’s offerings.
If the dataset is updated in the future, we will notify users who have provided their contact information.

Log in or Sign Up to review the conditions and access this dataset content.

Synthetic Luo (Dholuo) ASR Dataset

Synthetic Dholuo (luo) speech for automatic speech recognition (ASR), created by CLEAR Global. It contains about 775 hours of 24 kHz speech across 657,877 utterances, generated from LLM-written sentences by a text-to-speech (TTS) model fine-tuned on Dholuo Bible recordings.

This is a synthetic dataset. The sentences were written by a large language model and the audio was produced by a TTS model, not recorded from human speakers. The voice is a single male voice. Models trained on it should be evaluated on real speech before use.

Contents

  • data/train-*.parquet: 267 Parquet shards, one split (train).
  • audio: WAV audio, 24 kHz, mono, 16-bit PCM, embedded in the Parquet files.
  • text: normalised transcript, lowercased with punctuation removed. Use this for ASR training.
  • original_text: the sentence as generated by the LLM, with capitalisation and punctuation.
  • duration: clip length in seconds.
from datasets import load_dataset

ds = load_dataset("CLEAR-Global/Luo-Synthetic-ASR-Dataset", split="train")

How it was made

  1. Text. Dholuo sentences were written by Claude Sonnet 3.7: a corpus of 666,000 sentences (Table 3 of the paper). The linguist-evaluated sample of that corpus (1,600 sentences with readability and adequacy ratings) is published with the other languages as CLEAR-Global/Synthetic-Text, also on Mozilla Data Collective.
  2. Speech. A YourTTS model fine-tuned on Dholuo Bible recordings (BibleTTS-Luo, from open.bible) synthesised the sentences; this release holds the 657,877 clips produced for about 775 hours of audio. Model checkpoint: CLEAR-Global/YourTTS-Luo.

For the full method, evaluation and ASR results, see the paper: DeRenzi, Dixon, Farhi & Resch (2025), Synthetic Voice Data for Automatic Speech Recognition in African Languages, arXiv:2507.17578.

Limitations

  • Single male voice. The TTS model was fine-tuned on the single male Dholuo narrator in BibleTTS-Luo, so the whole corpus is one synthetic voice. No speaker metadata is included.
  • No natural acoustic variation. Real recordings contain tone, noise and speaking-style variation that synthetic speech lacks.
  • Model-induced bias. The text reflects the LLM's assumptions and may not capture Dholuo-speaking communities' culture or everyday language.
  • Bible-domain acoustic source. The voice was learned from Bible narration, so its pronunciation and prosody are shaped by that register rather than everyday speech.
  • Transcripts are the input sentences, not transcriptions of the audio. The text was generated first and then synthesised, so a clip's transcript shows what the TTS model was asked to say. Reviewers rated the text for readability in the paper's evaluation, but the audio was not checked against the transcripts clip by clip.

Related

Funding

This work was supported by the Gates Foundation. The conclusions and opinions expressed are those of the authors alone and shall not be attributed to the Foundation.

Citation

@misc{derenzi2025synthetic,
  title         = {Synthetic Voice Data for Automatic Speech Recognition in African Languages},
  author        = {DeRenzi, Brian and Dixon, Anna and Farhi, Mohamed Aymane and Resch, Christian},
  year          = {2025},
  eprint        = {2507.17578},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url           = {https://arxiv.org/abs/2507.17578}
}

License

Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0). See LICENSE. The synthetic text was generated with Claude Sonnet 3.7; the content is limited to non-commercial use and must comply with Anthropic's Terms of Service and Usage Policy.

Downloads last month
259

Paper for CLEAR-Global/Luo-Synthetic-ASR-Dataset