VIQ Solutions Submits Two Patent Applications for Methods Related to Training and Automated Selection of Domain Specific Language Models (DLSM)

VIQ Solutions, a global provider of secure, AI-driven, digital voice and video capture technology and transcription services, announces it has filed two provisional patent applications with the United States Patent and Trademark Office (USPTO). The patents augment the Company’s speech engine agnostic workflows, improving documentation accuracy and usability of documentation, for courts, insurance, law enforcement and media organizations across the globe.

AiThority Interview Insights: AiThority Interview with Arijit Sengupta, CEO and Founder at Aible

“Fewer edits equal greater efficiency and higher productivity as well as more capacity for faster turnaround orders, time savings, and cost reduction.”

The first patent application is directed to the systems and methods for training domain-specific Automated Speech Recognition (ASR) language models. The novel methods can be leveraged to form accurate speech recognition to easily manage nuances of domain-specific words, accented speech and languages, including mixed-language audio files, and document formatting requirements.

The second patent application is directed to the automated selection of domain specific ASR models. This patent aims to protect the various mechanisms used to load static or dynamic DSLMs. The static model is used based on a pre-assigned setting and the dynamic methods utilize the Goodness Score or Artificial Intelligence Natural Language Understanding (NLU) to determine the context and domain of the audio, followed by a second pass using a DSLM creating profound ASR improvements in domain-specific verticals.

Read More about AiThority Interview: AiThority Interview with Alex Mans, Founder and CEO at FLYR Lab

Both workflows utilize DSLMs to improve the quality of the draft transcript. DSLMs use a large database to create a baseline to create a new task-specific model that is being trained using the knowledge learned by the large language model. It leverages the knowledge learned during the training of the original model to improve the performance of the new model, resulting in increased efficiency, improved performance, and enhanced interpretability.

“DSLMs are one of many layers of aiAssist™ that supplies NetScribe with workflow automation and high-quality draft transcripts,” said Vahram Sukyas, Chief Technology Officer, VIQ Solutions. “We anticipate ASR improvements in domain-specific verticals by taking advantage of our large volume of production data to fine tune the DSLMs.”

The accuracy and usability of draft transcripts, FirstDraft™, is expected to increase and become even more competitive when enhanced with a series of “extra loops” in aiAssist workflow through VIQ trained DSLMs,” said Susan Sumner, President and Chief Operating Officer, VIQ Solutions. “Fewer edits equal greater efficiency and higher productivity as well as more capacity for faster turnaround orders, time savings, and cost reduction.”