Collecting Audio Data to train AI models for Deepfake Detection: Reality Defender x Voiceform

The Snapshot

‍Industry: AI & Cybersecurity

Use Case: Crowdsourcing diverse, high-quality voice data at scale to train AI deepfake detection models

Results at a Glance

• 500 interviews completed across 8 languages — replacing what would have been hundreds of individually scheduled video calls
• 50+ hours saved on data collection and participant coordination so far
• Higher quality datasets: built-in audio pre-processing and live playback cut technical issues and improved completion rates across three major dataset releases

‍

Challenge

Reality Defender secures communication channels against deepfake impersonations, enabling enterprises and governments to operate confidently. To maintain the highest standards in their anti-deepfake solutions, Reality Defender requires diverse and high-quality audio datasets. Typically, this process would involve scheduling one-on-one interviews, which are both expensive and time-consuming. Additionally, gathering such data across multiple languages and demographics presented challenges in scalability, technical compliance, and participant experience.

Solution

Reality Defender turned to Voiceform to collect authentic voice data from real humans at scale across eight languages, enabling the development of diverse and high-quality audio datasets for their AI models. By leveraging Voiceform, Reality Defender was able to conduct hundreds of interviews simultaneously, scaling what would typically be one-on-one interviews on a video call into an efficient, streamlined process. Voiceform’s conversational AI surveys played a pivotal role in achieving this scalability by creating an engaging and natural interaction for participants. Voiceform’s powerful and intuitive survey tools provided the foundation for:

• Audio Data Collection: Voiceform’s integrated audio tools enabled Reality Defender to capture high-quality audio data directly within surveys.
• Multi-Language Support: The multilingual capabilities in Voiceform allowed for customized workflows, reducing participant confusion and ensuring clarity across eight languages.
• Technical Compliance: Features such as audio pre-processing and live playback reduced technical issues and improved completion rates.

Implementation

Voiceform’s templating features, easy-to-use survey builder, and advanced capabilities enabled Reality Defender to:

1. Launch surveys in eight languages using Voiceform’s survey builder and translation capabilities, ensuring clarity and precision in every language.
2. Leverage Voiceform’s templating tools and collaboration features to streamline the survey design process, reducing time spent on repetitive tasks.
3. Utilize Voiceform’s AI-driven tools to enhance survey workflows and data quality, making it easier to manage complex datasets.
4. Rapidly prototype and deploy surveys with audio collection components.
5. Crowdsource diverse datasets for three major dataset releases.
6. Minimize time and resources spent on technical troubleshooting and participant onboarding.

Results

With Voiceform, Reality Defender achieved:

• Improved Data Quality: High-quality audio datasets with fewer technical issues.
• Enhanced Efficiency: Faster survey deployment and streamlined participant workflows.
• Expanded Reach: Successful data collection across eight languages, enabling Reality Defender to better serve their global clientele.

Curious to see a snapshot of Reality Defender's platform? See how it looks below.

‍

Ready to Collect Data for Training AI Models?

See how Voiceform can transform your AI model development.

Schedule a Demo | Try Free for 14 Days

Related Success Stories

📊 Prolific: 500 Qualitative Interviews in 72 Hours
🎓 Marist Poll: Adding Qual Insights to Quantitative Political Polls
💼 Boundless Markets: Scaling Qual Interviews While Saving 24 Hours/Week

‍