ALS Disease patient classification

The ALS voice dataset was collected at the Research and Clinical Center of Neurology and Neurosurgery (USC) and is designed to facilitate the study of speech impairments in patients with Amyotrophic Lateral Sclerosis (ALS). The dataset is comprised of 148 sustained vowel phonations, which include both pathological (ALS-affected) and healthy control (HC) voice samples. These phonations are critical in understanding speech degradation in ALS patients, as the disease affects motor neurons responsible for speech production. The collection of this data provides a valuable resource for machine learning models, acoustic analysis, and medical diagnostics related to ALS.

Dataset Composition

The dataset is balanced, containing 52% pathological voices (ALS patients) and 51% healthy voices (HCs). Each individual contributing to the dataset was asked to produce sustained phonation of the vowels /a/ and /i/ at a comfortable pitch and loudness, maintaining their voice as steadily and as long as possible. These vowels were specifically chosen as they provide significant phonetic markers in evaluating voice stability, intensity, and frequency variations, which can be affected by neuromuscular degeneration in ALS.

Challenges and Considerations

While the dataset provides valuable insights, there are some challenges and limitations:
• Device Variation: Since recordings were collected using different smartphones and headsets, there may be variations in audio quality and background noise.
• Limited Data Size: While 148 phonations provide a useful starting point, larger datasets are needed for more robust AI training and statistical analysis.
• Single Vowel Focus: The dataset focuses only on the vowels /a/ and /i/, which are useful but do not capture the full range of speech impairments seen in ALS.
• Speaker Variability: Individual differences in voice quality, pitch, and speaking habits may introduce variability, requiring careful normalization techniques during analysis.

Future Directions

Expanding and improving this dataset could lead to significant advancements in ALS diagnosis and speech therapy. Future efforts could include:
• Increasing the number of participants to enhance statistical power
• Expanding the phonetic range to include consonants and connected speech samples
• Using standardized recording devices to minimize technical variability
• Longitudinal data collection to track speech changes over time in ALS patients

This dataset represents an important step forward in ALS research, providing researchers with valuable voice recordings to study neurological speech disorders. With applications in AI, clinical diagnostics, and remote monitoring, this dataset has the potential to improve early detection, patient care, and disease progression tracking for ALS.