Utilizing AI to generate transcripts, creating broadcast-grade subtitles

The Challenge

To create broadcast-grade subtitles, several different challenges come into play. Professionals want to fine-tune the “spotting” rules of the captions, i.e. adjust the styling and timing. Mainstream broadcasters, such as the BBC and France Télévisions, have style guides close to 100 pages, to ensure maximum readability and accessibility.

The most obvious rules are about the timing and layout of the subtitles. But the more complex aspects are to determine the sweet spots for placing line breaks, and the requirement to synchronize subtitles with the rhythm of edit.

Limecraft’s unique Natural Language Processing (NLP) algorithm transforms the transcript into broadcast-grade subtitles, thereby taking into account the pre-configured spotting rules and information from the edit.

Subsequently, Limecraft built its platform as an extension of editing software. Using an Adobe panel, or as a plugin of Avid, journalists and editors get perfect subtitles delivered in less than a minute.

This level of automation and integration is of particular importance for creators of short-form video content. Otherwise, it would be a logistical nightmare to rely on subtitling companies where the turnaround time would be prohibitive.

The Solution

Benchmarks conducted by Limecraft's customers show that Speechmatics consistently delivers the best results for word error rate, speaker segmentation, capitalization, and punctuation. All these aspects are critical success factors given that the final deliverable is a subtitle file.

The unique combination of Speechmatics' outstanding ASR, alongside Limecraft's spotting algorithm, creates a wonderful solution for editors and journalists to create subtitles themselves. This drastically reduces the turnaround time and the cost compared to a manual captioning process.

The partnership between Speechmatics and Limecraft is a testament to the view that the supplier community has a moral duty to hide technical complexity and to mitigate implementation risks. A solid integration between the workspace application (in this case Limecraft) and the AI service provider (in this case Speechmatics) is essential for the reliability of the service and consistency of the results.

The Results

A typical broadcaster produces over 500 hours of content per month. Using Limecraft, the user saves 5,000 hours of manual subtitling work - that's the equivalent of 30 full-time staff or up to £100,000 per month. This allows the content creators to focus on what really matters – creating compelling stories.

This allows content producers to commit to their accessibility obligations, thereby reaching a much wider audience online (an estimated 50% of online content is played with captions on) and they drastically improve their SEO (clips with SRT files added are ranked significantly higher by Youtube).

Speechmatics first started working with Limecraft 6 years ago - punctuation had just launched and sensitivity was high. As you can see in this video (see above) from the bottom set of subtitles, Speechmatics has since made huge updates based on feedback and the results speak for themselves.

As technology continues to evolve, the Limecraft team know they can rely on Speechmatics as their ASR vendor to keep them at the forefront of AI innovation.