Improve transcription accuracy

Prerequisites

Voice transcription in Genesys Cloud converts speech into text, enabling a wide range of AI-driven use cases, including:

  • Real-time transcription for Agent Copilot
  • Post-call conversational intelligence
  • Content search
  • Summarization and insights
  • Sentiment and empathy analysis
  • Automated quality management
  • Data masking

Note: Bots and virtual agents use different transcription engines for recognizing customer utterances. For details, see About speech-to-text (STT) engines.

Accurate transcription is critical for these use cases. Errors can lead to incorrect agent assistance, faulty summaries, missed topics, or inaccurate scoring and sentiment analysis. Ensuring high transcription accuracy is essential before deploying advanced AI features.

Understanding transcription accuracy

Transcription accuracy is typically measured using Word Error Rate (WER).

Accuracy = 100% − WER%

For more information, see Word error rate.

Factors affecting accuracy

Several factors can influence transcription quality:

  • Audio and recording quality
  • Connectivity issues
  • The engine and training data used to build the model
  • Vocabulary, accents, and speaking styles

Diagnosing low transcription accuracy

If transcription accuracy is low, follow these steps:

  • Rule out systemic issues such as network, configuration, or service-related problems
  • Manually review and benchmark a sample of recordings
  • Identify frequently mis-transcribed vocabulary
  • Verify that queues and flows are configured with the proper dialect for accurate transcription

After addressing systemic or configuration issues, analyze errors by comparing transcripts to the audio.

Best practice: Review at least two hours of randomly selected recordings across the organization. Crowdsourcing input from supervisors or agents during this review can help log commonly mis-transcribed words for analysis.

Handling problematic words and phrases

Transcription issues are often concentrated on specific words or phrases. Recommended approaches include:

  • Add terms to dictionary management
  • Explore data donation

Add terms to dictionary management

Dictionary management improves recognition of business- or domain-specific terms such as brand names, acronyms, or internal terminology.

  • Terms can include example phrases, sounds-like guidance, and sensitivity boost values (1–10 on a logarithmic scale).
  • Boost values are only available through the API.

For more information, see Understand dictionary management.

Explore data donation

If dictionary management does not resolve issues, customers can donate recordings to help improve Genesys’ native voice transcription engine. Donated recordings:

  • Improve accuracy for unique accents, vocabulary, and speech patterns
  • Are redacted to remove personally identifiable information (PII) and company-specific terms
  • Follow a structured, secure workflow with Genesys to ensure privacy

Note: This process may take several months and involves direct collaboration with Genesys.

Roadmap for accuracy improvements

Genesys continuously updates:

  • Acoustic and language models
  • Underlying transcription engines

These updates drive ongoing improvements in transcription accuracy. For details on upcoming features, see Feature releases and communication. For additional questions about recordings or accuracy improvements, contact Customer Care.