Back to Blog
Product Updates3 min read

Voice AI Accuracy Improvements: What Changed in Q4

Vocable Team·

Our Q4 model update brings significant improvements to transcription accuracy and field extraction. Here's what changed and what it means for your forms.

Transcription Accuracy

Our updated speech recognition models deliver 15% better word error rate (WER) across all supported languages. The biggest improvements are in:

  • **Noisy environments**: Better performance in offices, clinics, and public spaces
  • **Accented speech**: More accurate recognition of regional accents and non-native speakers
  • **Technical vocabulary**: Improved handling of medical terms, legal language, and industry jargon

Smarter Field Extraction

The AI is now better at understanding context when extracting structured data from natural speech. Specific improvements include:

  • **Date parsing**: Handles relative dates ("last Tuesday"), partial dates ("March 2024"), and varied formats more reliably
  • **Name extraction**: Better at distinguishing first/last names in different cultural naming conventions
  • **Numeric values**: More accurate extraction of phone numbers, amounts, and measurements from conversational speech

Reduced Latency

Response time between user speech and AI follow-up questions has decreased by 200ms on average. Conversations feel more natural and responsive.

What You Need to Do

Nothing. These improvements are applied automatically to all forms. You may notice slightly different transcription results for edge cases, but overall accuracy should be noticeably better.

If you experience any regressions or unexpected behavior, please reach out to support@vocableform.com and we'll investigate immediately.