The dataset viewer is not available for this split.
Parquet error: Scan size limit exceeded: attempted to read 673188871 bytes, limit is 300000000 bytes
Make sure that
1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files
Error code: TooBigContentError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Abkhaz TTS — Alisa (Bagrat Shinkuba Fund)
A single-speaker Abkhaz (аҧсуа бызшәа, ISO 639 ab) text-to-speech corpus:
studio recordings of a single female narrator, Alisa, reading Abkhaz sentences,
paired with their transcripts. To our knowledge this is the first dedicated Abkhaz
TTS dataset on the Hub, filling a gap for a low-resource Caucasian language.
- Clips: 7,714
- Total audio: ~9.6 hours
- Audio: mono WAV, 48 kHz, 24-bit PCM
- Speaker: single female voice ("Alisa")
- Language: Abkhaz (Cyrillic script)
- Source: Bagrat Shinkuba Fund (Фонд Шинкуба) — literary sentences plus a spoken word/phrase list.
- License: CC0-1.0 (public domain dedication)
Columns
| column | type | description |
|---|---|---|
id |
int32 | Original index of the utterance in the source |
audio |
Audio (48 kHz) | Decoded waveform + sampling rate |
text |
string | Abkhaz transcript |
speaker |
string | Always Alisa |
Usage
from datasets import load_dataset
ds = load_dataset("Nart/abkhaz-tts", split="train")
print(ds[0]["text"])
print(ds[0]["audio"]["sampling_rate"]) # 48000
Notes
- Text is in the Abkhaz Cyrillic alphabet.
- One subset — the spoken word/phrase list at indices 6081–6460 (~379 entries) —
marks stress with a combining acute accent (´, U+0301) on a vowel. About half of
that subset (199 entries) carries a mark (e.g.
Мшыбзи́а!"Hello!"). No other part of the corpus uses stress marks, so strip U+0301 if you need fully uniform transcripts.
License
Released under CC0-1.0: no rights reserved, free for any use including commercial. Attribution to the source collection (Bagrat Shinkuba Fund) is appreciated but not required.
- Downloads last month
- 77