Data
What we offer.
A Tiv-English parallel text corpus and studio audio recorded from it. One clip per sentence, each with its text, translation and metadata.
Parallel text corpus
Tiv-English, 7,000+ sentences, 27 categories.
Sentences cover everyday and specialist language, from healthcare and government to idioms and proverbs. Each Tiv sentence is paired with an English translation.
Categories
- [TODO: list all 27 category names from the founder]
Studio audio
One clip per sentence.
Each recording is paired with its Tiv text and English translation. Audio is released category by category. It is currently available as samples.
Format
- File format
- [TODO: confirm, e.g. WAV]
- Sample rate
- [TODO: confirm]
- Channels
- [TODO: confirm mono]
Metadata
Every clip carries its provenance.
Each clip comes with metadata so you can audit where it came from and how it was checked.
Fields per clip
- Speaker ID
- Identifies the speaker for each recording.
- Recording setup
- Studio, microphone and sample rate.
- Transcript
- Source of the transcript and its verification status.
- Category
- The category the sentence belongs to.
[TODO: founder to confirm the final metadata field names]
Intended uses
Built for voice and language models.
- Text-to-speech
- Automatic speech recognition
- Machine translation
- Evaluation
Licensing
Commercial, non-exclusive licences.
Terms are on request. Tell us what you are building and we will discuss the right licence.
The open samples are licensed CC BY-NC 4.0 and are for non-commercial evaluation only. Commercial use of any part of the data needs a licence.