Custom Whisper model, 99% transcription accuracy

The Challenge

After launching an earlier transcription service, the team found that certain recordings — handheld audio from community events rather than professional-grade audio — couldn't be transcribed by hyperscaler services due to background noise and other distractions.

Our Solution

A custom Whisper-based transcription service tuned to filter out noise and distraction, reaching 99% transcription accuracy while cutting production time by roughly 75% across the client's 100+ supported languages.

Impacts

• 99% transcription accuracy
• ~75% reduction in production time
• Dramatically faster time to publish
• Standardization across all supported languages