Generative AI for UNE 153010-compliant subtitles with 70–80% less manual editing

Growing demand for accessible, multichannel audiovisual content (TV, OTT platforms, social media, corporate content) was putting increasing pressure on the client's subtitling teams. Although they were already using automatic transcription tools, the resulting files had:

• Language and punctuation errors.
• Timing mismatches with the spoken audio.
• Recurring breaches of the UNE 153010 standard in line length, segmentation, layout, and formatting.

All of this meant heavy manual review, tight delivery deadlines, and a growing risk of compliance inconsistencies across projects and clients.

The client, with a clear focus on innovation and accessibility, set out to turn this challenge into an opportunity to modernize its value chain: cut editing time without compromising on quality, ensure regulatory compliance systematically, and free up subtitling professionals for higher-value work.

To tackle this challenge, the approach was structured in four phases:

1. Assessment and functional design

• Analysis of existing subtitling workflows, from automatic transcription to final delivery to the client.
• Identification of friction points: the most frequent types of errors, editing time per minute of video, and variability between subtitlers.
• Translation of the UNE 153010 standard into a set of operational rules an automated system can apply (text length and distribution, minimum and maximum on-screen times, segmentation by units of meaning, etc.).

2. Solution design and high-level architecture

• Definition of a specialized application that takes in subtitle files produced by automatic transcription engines.
• Selection and integration of a state-of-the-art generative AI model, running in a secure cloud environment and able to understand linguistic context and apply formal and regulatory rules.
• Design of an interface and workflow built around the subtitler, so the tool fits naturally into existing processes.

3. Development of key features

The application was built around three core capabilities:

• Initial subtitle cleanup: starting from the file generated by automatic transcription, the AI identifies and corrects spelling and grammar errors, detects timing and on-screen duration irregularities, and adjusts the formatting and layout of the text to align with UNE 153010.
• Continuous verification and iterative correction: the model includes a self-assessment mechanism that checks the file against the standard; whenever it finds a deviation, the AI itself proposes and applies further corrections, iterating until it reaches 100% compliance with the defined criteria.
• Delivery of optimized files: the output is subtitles that are practically ready to publish, which the subtitler only needs to review and approve. The workflow shifts from "correcting line by line" to "reviewing and fine-tuning an already optimized base."

4. Pilot, calibration, and rollout

• A pilot with different types of content (fiction, news, documentaries, corporate content) to calibrate the sensitivity of the rules and the model's configuration.
• Fine-tuning of parameters (segmentation, reading times, language style) and definition of templates by client or channel.
• Gradual rollout across the client's operations, supported by subtitler training and changes to internal protocols to maximize adoption of the tool.

The solution was designed with an adaptable architecture that makes it possible to add other standards or the specific requirements of international clients while keeping a core based on generative AI and configurable business rules.

Implementing the application has delivered measurable, differentiating impact on the client's business:

Greater operational efficiency

• Manual editing time reduced by around 70–80%, significantly shortening delivery times per project.
• An estimated increase of up to 2–3 times in the minutes of video processed per subtitling team, at constant capacity.

Systematic regulatory compliance

• Automated alignment with the UNE 153010 standard, ensuring consistent subtitle formatting, segmentation, and timing.
• A very marked drop in internal quality reviews tied to formal non-compliance, reducing rework and reputational risk.

Higher quality and fewer errors

• A significant decrease in language errors and timing mismatches in final deliveries, with a more uniform level of quality across projects and teams.
• The model's iterative reflection (self-assessment and correction over several passes) reduces residual error margins and stabilizes the quality standard.

A boost to accessibility and brand positioning

• By making it easier to create high-quality, standards-compliant subtitles, the solution improves the experience of people with hearing disabilities and of audiences who consume subtitled content.
• The client strengthens its position as a go-to strategic partner for audiovisual accessibility and regulatory compliance in its market.

Technology modernization and competitive advantage

• A generative AI-based tool that turns a traditionally manual process into a workflow supported by emerging technologies.
• Adaptable customization to work with other standards and the specific requirements of large clients, creating a value proposition that competitors without similar capabilities would find hard to replicate.

Overall, the project shows how integrating generative AI into industrial subtitling processes makes it possible to tackle complex challenges of efficiency, quality, and regulatory compliance, delivering a tangible return in productivity and reinforcing the client's role as a key player in the digital transformation of the audiovisual sector.

Spanish Business Awards 2025

Best Digitization Consultancy Firm 2025 - Madrid