resproj007's picture
Update README.md
07ed881 verified
metadata
license: mit
language: en
tags:
  - pathological-speech
  - speech-synthesis
  - tts
  - voice-conversion
  - healthy-control
  - torgo

Torgo Healthy Female Dataset (Updated)

Overview

This dataset contains healthy control speech samples from a female speaker (FC02) in the TORGO corpus, prepared for pathological speech synthesis research.

Speaker Information:

  • Speaker ID: FC02
  • Corpus: TORGO
  • Gender: Female
  • Speech Status: Healthy Control

Dataset Statistics

  • Total Samples: 800
  • Total Duration: 0.63 hours
  • Sampling Rate: 24,000 Hz
  • Format: Audio arrays with transcriptions

Training Split

  • Samples: 700
  • Duration: 0.55 hours
  • Avg Duration: 2.9s
  • Duration Range: 1.6s - 7.5s
  • Avg Text Length: 13 characters

Test Split

  • Samples: 100
  • Duration: 0.08 hours
  • Avg Duration: 2.9s
  • Duration Range: 1.9s - 6.3s
  • Avg Text Length: 14 characters

Loading the Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("your-username/torgo_healthy_female")

# Access train and test splits
train_data = dataset['train']
test_data = dataset['test']

# Each sample contains:
# - 'audio': {'array': numpy_array, 'sampling_rate': 24000}
# - 'text': str (normalized transcription)

# Example usage
sample = train_data[0]
audio_array = sample['audio']['array']
transcription = sample['text']
sampling_rate = sample['audio']['sampling_rate']

Direct Training with Transformers

from transformers import Trainer
from datasets import load_dataset

# Load and use directly with Trainer (no preprocessing needed)
dataset = load_dataset("your-username/torgo_healthy_female")
trainer = Trainer(
    train_dataset=dataset['train'],
    eval_dataset=dataset['test'],
    # ... other trainer arguments
)