Improving transcription accuracy

This article is a guide on the possible ways of improving the recognition rate of the transcriptions. Speech transcription engine uses a language model that describes the spoken language acoustically and linguistically. There are two options for improving the generic language model, language and accent customization (LAC) and phonetic boosting.


Language and accent customization (LAC) 

The language and accent customization is a professional service that focuses on the acoustic characteristics of the language model. It takes into consideration specific environmental and spoken elements, such as background noises, organization scripts or language patterns. For this process, it is necessary to collect audio.  

Please contact us for further details.

Phonetic boosting

Phonetic boosting is a fine tuning customization option for the linguistic characteristics. It helps to increase the recognition rate if the transcription system needs improvements with company-specific words. The following actions can be done:

  • Expanding the language model vocabulary
    Add terms and phrases that are unique to your business or to an ongoing campaign to boost their recognition by the Speech Engine.
  • Boosting recognition of existing terms and phrases 
    The most common cause for poor recognition of a term or phrase is the fact that its probability is higher in your field. Changing the weight of an existing term or phrase increases its probability, changing its impact on the Speech Analytics results.
  • Suppressing a competing phrase
    An incorrect transcription of a term or phrase causes it to compete with the “correct” transcription. With Phonetics Boosting, you can suppress the incorrect phrase by decreasing its band probability.
  • Supplementing pronunciations
    If a term has several pronunciations, adding all the pronunciation variants boosts recognition

Please contact the support service for assistance in phonetic boosting.