demeleww/Amharic_tokens
Domain:
natural language processingRecord type:
dataset
Creator:
dem
Host:
Pre-extracted HiggsAudioV2 audio tokens for training OmniVoice Amharic TTS.
Downloading these tokens (~648 MB, ~2 min) skips the 30+ minute token extraction step (steps 1-3 of the training pipeline).
Stat
Value
Total samples
81,731
Total audio
~331 hours
Avg sample duration
~14.6 seconds
Audio shards
164 (.tar files)
Text shards
164 (.jsonl files)
Samples per shard