Improving productivity or operational efficiency gains using smart transcription is one of the key driver for any digital transformation effort in the Healthcare and Petcare industry. Any noticeable improvements in these parameters results in direct bottom-line gains, and enabling the professionals to do more of what they are good at – seeing more patients.
The current implementations EMR or PMS (Practice Management Software) applications that deal with the day-to-day operations of a medical practice heavily involve the typical graphical user interfaces (GUI), that mandate the users (Medical Professionals) to use the Keyboard and Pointing devices actively. It becomes a cumbersome and time consuming activity every time, when they have to remove their gloves, use the GUI devices and return back to dealing with the patient, more so in the case of veterinarians who treat the pets.
The business problem involved using a Voice interface not just for transcription of exam findings, plan and assessments, but converting the captured unstructured text into structured medical records for storage, thereby saving significant amount of documentation time for the medical professionals.
One of the major (make or break) project requirement was to achieve very high levels of transcription accuracy, closer to human comprehension levels, with complex medical vocabulary.
Voice interface technologies help solve two major design problems – recognition and understanding.
Recognition is the part of technology that applies to converting Speech to Text, while Understanding refers more to the semantics and the ability provided to computers to comprehend and attach context specific meaning to the spoken words to take action.
While the vendor Speech and Language Understanding Services are good with regular English vocabulary, the biggest challenge for enterprises to achieve reliable and predictable accuracy levels is in handling specialized vocabulary in their own domain, its usage in day-to-day operations by their staff.
How did we achieve the desired accuracy levels in the smart transcription?
- Usage of Domain Specific Vocabulary for Speech to Text and Language
Understanding for Entity Extraction
- Specialized NLU post-processing using Heuristic as well as Probabilistic techniques
for Entity Association
- Semantic Graph for knowledge corpus
- Continuous retraining of ML models based on usage and error scenarios.
- Implemented a metrics based approach to mature the models.
- Simple web application tools for Domain Experts to collect, monitor and mature
- language accuracy with training workflows
- Comprehensive error-rate dashboard to measure accuracy
- With all the voice data processed on the device and behind the firewall, the application is intrinsically HIPAA and GDPR complaint
We are a Conversational AI startup, driven by passion to build great software solutions for our customers. As machines become increasingly proficient in the language we speak, we see Conversations, in Voice and Text, as the primary mode of providing and consuming services. We help our customers build the next generation Conversational AI Applications that primarily use Language as the Interface. Our platform and applications, primarily offered as a low-code offering and our belief in modern, lean software engineering principles helps our customers with higher effectiveness and time to market.