Evidence-Based Recommendations for Tablet Recordings From the Bridge2AI-Voice Acoustic Experiments

Area of Science:

• Speech and Audio Processing
• Biomedical Engineering
• Artificial Intelligence

Background:

• Voice data collection is crucial for advancing voice artificial intelligence (AI) research.
• Establishing best practices for data collection is essential for reliable AI models.
• Investigating accessible recording methods is key to large-scale data acquisition.

Purpose of the Study:

• To evaluate the efficacy of iOS and Android tablets with and without low-cost headset microphones for voice data collection.
• To compare acoustic measures from tablet-based recordings against research-grade instrumentation.
• To determine the suitability of consumer-grade devices for producing research-quality voice data.

Main Methods:

• Sustained vowel recordings from 24 participants (typical and disordered voices) were used.
• Recordings were made using a research-standard setup and two popular tablets (built-in mics and headset mics).
• Acoustic measurements were compared across different recording setups and distances.

Main Results:

• Tablets with headset microphones placed close to the mouth (2.5-5 cm) showed strong correlations (r > 0.90) with the research standard.
• No significant differences in vocal frequency and perturbation measures were found with headset microphones.
• Tablet built-in microphones at typical distances (30-45 cm) exhibited significant variability and poorer correlations.

Conclusions:

• Smartphones with low-cost headset microphones are adequate for large-scale voice data collection.
• Headset microphones ensure consistent recording distance and reduce background noise, improving data quality.
• This approach supports the Bridge2AI-Voice Consortium's recommendations for voice AI research.