Syncwords: Leveraging Machine Learning to Meet an Accessibility Challenge

Syncwords was created as an extension of Frameweld’s experience with closed captioning. Driven by our need to caption our clients’ videos from transcripts, we decided to build a product that would not only help us generate captions quickly and accurately, but could also be used by other companies and individuals to achieve similar results.

Creating closed captions is typically a labor-intensive process, and can become very expensive, as well. Automated time-stamping of captions can be challenging on long duration media files, especially if the audio has background noise, music and multiple speakers such as found in broadcast media.

"Syncwords' creation and growth is a result of listening and acting on our customers' needs and demands for high-quality captions." Sam Cartsos
Co-founder, Frameweld

The Solution

Using sophisticated machine learning algorithms and methods involving Deep Neural Networks, we developed Syncwords into a sophisticated tool to generate captions rapidly and accurately from media and transcript. Besides time-stamping accuracy, high quality captions are also segmented or chunked appropriately for better readability and comprehension by the viewer. While this task is traditionally done by human captioners, Syncwords automated it using natural language processing (NLP) and machine learning based algorithms.

How it's done

SyncWords’ chunking performance is very close to human performance under the demanding standards required by the leading post-production and captioning companies. While this task is traditionally done by human captioners, Syncwords automated it using natural language processing (NLP) and machine learning based algorithms. SyncWords’ chunking performance is very close to human performance under the demanding standards required by the leading post-production and captioning companies.