resproj007/sesame_pathological_synthetic_data

Combined Fine-tuned Sesame CSM 1B Synthetic Speech Dataset

Data Format

Each sample contains:

• audio: Audio array with sampling_rate (24kHz)
• text: Original transcript text
• speaker_id: Speaker identifier (F04, M02, FC02, MC01, F02, M04, 211, 4014)
• corpus: Source corpus (TORGO, UA-Speech, LibriSpeech)
• condition: Speaker condition (Dysarthric, Healthy)
• model_name: Fine-tuned model name
• model_type: "sesame_csm_1b_adapter"
• base_model: Base model used (unsloth/csm-1b)
• adapter_path: HuggingFace adapter path
• duration: Audio duration in seconds
• sample_rate: Audio sample rate (24000)
• sample_index: Sample index number
• audio_file: Original audio filename
• generated_at: Generation timestamp

Usage Example

from datasets import Dataset
import pandas as pd

# Load combined dataset
dataset = Dataset.from_parquet("sesame_tts_synthetic_combined.parquet")

print(f"Total samples: {len(dataset)}")
print(f"Speakers: {set(dataset['speaker_id'])}")

# Filter by model type
sesame_samples = dataset.filter(lambda x: x['model_type'] == 'sesame_csm_1b_adapter')
print(f"Sesame CSM samples: {len(sesame_samples)}")

# Filter by condition
dysarthric_samples = dataset.filter(lambda x: x['condition'] == 'Dysarthric')
healthy_samples = dataset.filter(lambda x: x['condition'] == 'Healthy')

print(f"Dysarthric: {len(dysarthric_samples)}, Healthy: {len(healthy_samples)}")

# Play sample
sample = dataset[0]
from IPython.display import Audio
Audio(sample['audio']['array'], rate=sample['audio']['sampling_rate'])

# Convert to DataFrame for analysis
df = dataset.to_pandas()
print(df.groupby(['corpus', 'condition'])['duration'].agg(['count', 'sum', 'mean']))