metadata

dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: text
      dtype: string
    - name: age
      dtype: string
    - name: gender
      dtype: string
    - name: client_id
      dtype: string
  splits:
    - name: train
      num_bytes: 342225811.1
      num_examples: 9300
    - name: test
      num_bytes: 207494311.74
      num_examples: 5818
    - name: eval
      num_bytes: 199669352.112
      num_examples: 5856
    - name: generated
      num_bytes: 20007480833
      num_examples: 100000
  download_size: 19300182564
  dataset_size: 20756870307.952
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
      - split: eval
        path: data/eval-*
      - split: generated
        path: data/generated-*
license: cc0-1.0
language:
  - hy
pretty_name: Hy-Generated

Hy-Generated Audio Data with CV20.0

This dataset provides Armenian speech data consisting of both real and generated audio clips.

The train, test, and eval splits are derived from the Common Voice 20.0 Armenian dataset.
The generated split contains 100,000 high-quality clips synthesized using a fine-tuned F5-TTS model, covering 404 equal distribution of synthetic voices.

📊 Dataset Statistics

Split	# Clips	Duration (hours)
`train`	9,300	13.53
`test`	5,818	9.16
`eval`	5,856	8.76
`generated`	100,000	113.61

Total duration: ~145 hours

🛠️ Loading the Dataset

from datasets import load_dataset

dataset = load_dataset("ErikMkrtchyan/Hy-Generated-audio-data-with-cv20.0")