Faster Session Notes with Speech-to-Text AI for Healthcare Providers

View All Reference Kits

Background

Substance abuse disorders, personality disorders, depression, or addictive behaviors are the most common health conditions affecting one in every five individuals in the US.¹ Additionally, it has been found that about one in four Americans experience a mental disorder each year.² Only around 40 percent of US adults with mental illness received treatment in 2019 partly due to a shortage of mental health providers.³ The National Council for Mental Wellbeing predicts this shortage can exceed over 15,000 needed clinicians in the next couple of years.⁴

Mental health providers are required to document their sessions using progress notes, which provide an outline of the basic information about what occurred during a session.⁵ While patient progress notes have various uses, they can consume about 35% of the productivity of the healthcare providers.⁶ Healthcare providers can save time by recording session notes using speech-to-text AI algorithms that are optimized for performance with Intel® software products.

Solution

In collaboration with Accenture*, Intel developed this AI transcription reference kit that may be incorporated into an application. The kit assists the healthcare providers to save time in recording session notes. The kit features speech-to-text AI algorithms that when paired with Intel software products may help your applications complete the speech-to-text transcription of their recordings faster. This reference kit includes:

• Training data
• An open source, trained model
• Libraries
• User guides
• Intel® AI software products

PyTorch* is used to perform generative adversarial network (GAN) model training. GAN is a type of generative modeling that performs unsupervised learning on the audio samples. The trained model is then used with the audio input to transcribe it to phonemic text.

Intel® Optimization for PyTorch* accelerates the PyTorch performance on Intel hardware with minimal code changes, which results in a faster transcription output.

At a Glance

• Industry: Healthcare
• Task:
  • Data preparation of speech and text data features.
  • Preprocessing the unstructured voice samples into segmented unlabeled audio samples.
  • Training on top of the preprocessing features to map the representations to phonemes. Loop in new data to identify the best model.
  • Run inference with best model
• Dataset: Speech voice samples
• Type of Learning: Unsupervised, transfer learning
• Models:
  • Pretrained PyTorch model (Wav2vec_small) for data preparation
  • Trained GAN Model with PyTorch for inference
• Output: Phonemic text
• Intel® AI Software Products:
  • Intel Optimization for PyTorch

Technology

Optimized with Intel® AI Software Products for Better Performance.

Intel Optimization for PyTorch is made available as part of the Intel® AI Analytics Toolkit (AI Kit) that provides a comprehensive and interoperable set of AI software libraries to accelerate end-to-end data science and machine learning workflows.

Performance was tested on Microsoft Azure* Standard_D8_V5 using 3rd generation Intel® Xeon® processors for optimized performance.

Benefits

As the number of Americans experiencing mental disorders increases and the availability of mental health providers decreases, streamlining steps in a provider's daily practice becomes more important.

This AI transcription reference kit offers the healthcare organizational developer the opportunity to improve the productivity of their healthcare providers by incorporating these speech-to-text AI algorithms, which are optimized to run even faster with Intel software products.

Download Kit

Additional Resources

AI Tools

Intel Optimization for PyTorch

Developer Resources from Intel and Accenture

References

1. Mental Health by the Numbers. National Alliance on Mental Illness. (2021). Nami.org; https://www.nami.org/mhstats
2. Mental Health Disorder Statistics, Johns Hopkins Medicine, 2022, https://www.hopkinsmedicine.org/health/wellness-and-prevention/mental-health-disorder-statistics
3. National Projections of Supply and Demand for Selected Behavioral Health Practitioners: 2013-2025, Health Resources and Services Administration Bureau of Health Workforce National Center for Health Workforce Analysis. (2016). https://bhw.hrsa.gov/sites/default/files/bureau-health-workforce/data-research/behavioral-health-2013-2025.pdf
4. Nearly 80% of National Council for Mental Wellbeing Members Say Demand for Treatment Has Increased over the Past Three Months. National Council for Mental Wellbeing. Retrieved August 26, 2022, https://www.thenationalcouncil.org/news/nearly-80-of-national-council-for-mental-wellbeing-members-say-demand-for-treatment-has-increased-over-the-past-three-months/
5. Psychotherapy Notes. Good Therapy. https://www.goodtherapy.org/blog/psychpedia/psychotherapy-notes
6. Time Spent on Dedicated Patient Care and Documentation Tasks Before and After the Introduction of a Structured and Standardized Electronic Health Record. National Library of Medicine, NIH. 2018 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5801881/