Datasets:
Download DATA_USE_TERMS.md from Chithekitale/Chichewa_TTS_ASR: direct link, hf CLI and curl.
- Browser
- Download file 4.31 kB
-
https://huggingface.co/datasets/Chithekitale/Chichewa_TTS_ASR/resolve/main/DATA_USE_TERMS.md
- Command line
-
hf download hf://datasets/Chithekitale/Chichewa_TTS_ASR/DATA_USE_TERMS.md
-
curl -L -o DATA_USE_TERMS.md https://huggingface.co/datasets/Chithekitale/Chichewa_TTS_ASR/resolve/main/DATA_USE_TERMS.md
Chichewa Speech Corpus
Data Access and Use Terms
Access model — Controlled access for approved non-commercial research
Dataset steward — Dr Stanley Ndebvu, Mzuzu University
Contact — ndebvu.s@mzuni.ac.mw
Dataset version and repository record — Version 1.0 and
Effective date — November 20, 2026
Before requesting access: Read these terms and explain how you intend to use the corpus. The audio contains recognisable human voices. Access is granted to individual researchers for an approved project.
1. Requesting access
Provide your name, affiliation, institutional email address, research purpose, and requested access period. Describe how you will store and process the files, including any external computing or storage services you plan to use. Each collaborator who needs access must submit a separate request.
2. Permitted research
Approved users may download and process the corpus for the non-commercial project described in their approval. Projects may include Chichewa text-to-speech, automatic speech recognition, or linguistic analysis. Obtain separate written authorisation before making a material change to the research purpose or using the corpus for a commercial purpose.
3. Protecting the speakers
Removing names does not prevent listeners from recognising a speaker’s voice. Do not use the corpus to identify, contact, or profile speakers, including by linking recordings to other sources. Do not use it to impersonate a speaker or generate deceptive audio. Research specifically intended to reproduce a participant’s recognisable voice requires separate written approval and must be within the scope of that participant’s consent.
4. Sharing files and research outputs
| Material | Sharing conditions |
|---|---|
| Dataset files, copies, and access credentials | Keep these within the approved access arrangements. Do not share credentials or redistribute recordings, transcripts, or other dataset files. |
| Aggregate findings and research code | You may publish these if they contain no dataset material and do not disclose speaker identities. |
| Audio samples, transcript excerpts, and speaker embeddings | Obtain written approval before sharing these outside the approved project. |
| Models, checkpoints, demos, or generated audio that could reproduce a recognisable participant’s voice or expose dataset material | Obtain written approval before making these available outside the approved project. |
Use external computing or storage services only as covered by your approved data-handling arrangements. Do not upload the corpus to public repositories or to services that may retain or reuse it for their own purposes.
5. Security, incidents, and deletion
- Secure the files. Store them securely and restrict access to approved individuals.
- Report incidents promptly. Notify the dataset steward of suspected loss, unauthorised access, disclosure, or misuse, and take reasonable steps to contain it.
- Delete copies when access ends. Stop using the corpus and delete copies under your control when the approved access period ends, the project concludes, or access is withdrawn, whichever comes first. If a legal or institutional obligation requires retention, notify the steward and keep the required copy securely restricted.
6. Citation and reporting
Cite the creator, dataset title, version, and DOI or repository record. Cite Stanley Ndebvu’s University of Botswana doctoral thesis when describing the original collection methods or experiments.
When reporting model performance, identify the evaluation split and state whether speakers overlap across training and test sets. The corpus captures scripted read speech from five professional broadcasters in Malawi. Claims about other speakers, regional varieties, or conversational speech require separate evidence.
7. Acceptance and withdrawal
By accepting these terms and receiving access, you agree to follow them and any conditions in your written approval. The dataset steward may suspend or withdraw access for a breach or when required to protect participants or comply with applicable ethics requirements. The restrictions on redistribution and speaker identification continue after access ends.