Improve transcription accuracy

Within Genesys Cloud speech and text analytics (STA), voice transcription provides the value of unlocking the meaning of interactions by turning speech into text. STA also frees supervisors and quality managers from listening to an interaction recording.

Voice transcription lays the foundation for further interaction analysis. For instance, content search, topic spotting, sentiment analysis, empathy, and unprofessional behavior are all built on transcription analysis. 

Accurate transcription provides useful insights. Poorly transcribed calls force a supervisor to listen to the recording for missed topics and for sentiments that maybe scored incorrectly. Better accuracy improves all STA capabilities. 

Understand transcription accuracy

Speech transcription accuracy is typically understood using the Word Error Rate (WER) measurement. The accuracy is represented by a negative WER value (for example, 100% – WER%). For more information, see Word error rate.

Factors that can affect the transcription accuracy

These factors affect the transcription accuracy:

• Audio quality
• Recording quality
• Connectivity issues
• Configuration of transcription services
• Engine used for training and transcription
• Training data used to build model
• Vocabulary used in interactions

Low transcription accuracy

When transcription accuracy is low, perform the following steps to ensure that no underlying issues exist to negatively affect transcription.

1. Rule out systemic issues.
2. Manually review, transcribe, and benchmark a sample of the recordings.
3. Check for vocabulary that is regularly transcribed incorrectly.

After you perform these steps, verify that the configuration is correct. Queues and flows must be set to the proper dialect for accurate transcription functionality.

After you rule out configuration and any underlying issues, analyze the extent of the transcription errors. To perform this analysis, collect sample recordings and compare the generated transcripts to the manually transcribed transcripts. To reach a high recording accuracy confidence level, this process may require approximately two hours of transcribed recordings. The two hours of transcribed recordings should include a random assortment of interactions, rather than a selection of seemingly problematic interactions.

For more information about exporting recordings from Genesys Cloud, see About the AWS S3 recording bulk actions integration.

Problematic words and phrases 

Often, the problem does not reside in the overall transcription; instead the issue is about specific words that transcribe inaccurately.

The following list represents the methods that you can perform to improve problematic words and phrases.

• Add terms and phrases using Dictionary Management
• Add phrases to the topics definition
• Use Extended Voice Transcription Services
• Data donation
• Words discovery

Add terms and phrases via Dictionary Management

Dictionary management provides a means of improving recognition for business or domain-specific terms. Specific brands, words, or acronyms are transcribed based on the organization’s specifics. This feature allows customers to add terms to the dictionary, which enhances the transcription service’s likelihood of recognition. For more information, see Understand dictionary management.

With dictionary management, specify organization-specific terms, provide example phrases, identify similar-sounding terms, and assign sensitivity boost for voice transcription. Boost values range from 1 to 10 and increase the term’s identification likelihood on a logarithmic scale. Boost values are only available with the API.

Add phrases to topics definition

Topic spotting allows companies to spot specific topics within interactions. A topic encapsulates a business-level intent detected within interactions. By adding problematic words and phrases as a Topic and ensuring queues are configured properly, you can bias the transcription engine toward identifying these phrases. For more information, see Work with a topic.

Use the Extended Voice Transcription Services (EVTS)

EVTS uses the Azure STT integration to transcribe voice interactions. The integration directly integrates into Genesys Cloud so that the transcript appears in the Interactions Detail page just like any other transcribed transcript.

Depending on the described dialect, transcription accuracy may be greater using EVTS.

For more information, see Voice transcription – What is the difference between Genesys Cloud Voice Transcription and Extended Voice Transcription Services. For more information about EVTS supported languages, see Genesys Cloud supported languages. 

Data donation

When you include recordings with manual transcriptions, you can retrain your host language transcription engine with company-related terminology and improve accuracy. Data associated with recordings with manual transcripts improves overall accuracy for all dialects. It may take several months for this data to process, and for the dialect model to be retrained. A donation of 200 hours or more can lead to an improvement in accuracy.

Roadmap accuracy improvements

Genesys continually updates acoustic and language models, and also the underlying transcription engines used for transcription. This process leads to continuous accuracy improvements. For more information about an upcoming feature release, see Feature releases and communication. For more questions about recordings with manual transcripts data, and for accuracy improvements, contact Customer Care.