Vendor Content.

How Broadcasters Can Simplify Closed Captioning For Complex Networks

This article from captioning technology leader Ai-Media explores how combining automatic captioning with human curation creates high quality captions for complex broadcast networks, providing a highly scalable and cost-effective solution that supports multiple languages.

Broadcasters around the globe are reaping significant benefits from the rapid advances taking place in the closed captioning space. Like many areas of media technology, large-scale shifts are unfolding for captions, resulting in broadcast audiences gaining even greater accessibility and enjoyable experiences.

The rapid evolution of caption systems is creating many new efficiencies for broadcasters. These efficiency upgrades are essential, since captioning talent and technologies must meet ever-higher standards for connectivity, interoperability, security, ease-of-use, affordability, and accuracy. The latest improvements also deliver better flexibility to caption workflows.

Ensuring accessibility for those with hearing loss has always been a major driver for broadcasters, but their incentives to ensure quality captioning extend well beyond compliance. Research shows that captions lead to improved engagement and longer average watch times. While advances in automatic caption translation streamline distribution to new markets, enabling new revenue streams.

How Combining ASR And Human Captioning Benefits Broadcasters

As broadcast networks scale up, they face increasingly complex choices regarding captioning. Automatic speech recognition (ASR) solutions like Ai-Media’s Lexi, also known as automatic captioning, provide a simplified choice. These solutions lend themselves well to the efficiency of IP-based workflows and cloud software-driven environments. Lexi leverages an AI-driven Topic Models system – otherwise known as a custom dictionary – to ensure that it accurately displays highly customized vocabularies and phrases.

However, even the most advanced automatic captioning might not always meet desired accuracy levels for certain types of programming like live news or sports, which often include names or terms that typical speech-to-text systems can’t recognize. In these cases, premium human captioners provide the highest level of quality. Human captions also typically come with a higher level of support, such as for scheduling or troubleshooting, than is available through software alone. The drawback to human captioning is that it carries a significantly higher cost per hour than ASR.

A hybrid approach can help broadcasters solve this dilemma. Ai-Media’s Smart Lexi is an advanced offering of Lexi that combines machine learning automation with expert human curation. It builds on Lexi’s high accuracy, automation, and affordability by adding comprehensive human management and support for more complex captioning needs.

This is an optimal method for broadcasters seeking live captions that achieve higher accuracy than out-of-the-box automated captioning solutions, or at a lower cost than premium human captioning services. By layering these techniques and technologies, in-depth research performed by content teams can be leveraged on live broadcasts and streaming. Captioning professionals build expanded Topic Models by referring to internal databases and customer-provided documentation.

This hybrid approach is very cost-effective, providing the quality of human captioning at approximately half the cost of that service. It gives users the ability to create regular reports confirming accessibility compliance standards are met. 24/7 monitoring is also often included to ensure optimal caption delivery, as is the ability to constantly update custom vocabularies. Onsite captioning management can also be deployed for particularly complex projects. This system is in use by extensive content networks and global sports competitions, with thousands of programming hours and dozens of live streaming events to live caption.

A Streamlined Signal Path

The key ingredient that makes Lexi and Smart Lexi such reliable, secure and easy-to-manage solutions is Ai-Media’s globally adopted iCap captioning and subtitle delivery network – which is the largest of its kind in the world. iCap provides broadcasters with a simple pathway to integrating Smart Lexi.

As in typical captioning workflows, the caption stream – either human or ASR – is converged with the video stream. The IP-based captioning network streamlines the management of communications between protocols and people.

With a hybrid captioning system such as Smart Lexi, broadcast engineers receive extremely comprehensive human support. It consists of five main components:

Initial setup – commissioning of the system

Custom dictionary management – the ASR’s AI engine is continuously updated to maximize caption accuracy

Live support for job scheduling and management – which typically includes priority tech support for any issues

Constant monitoring – caption quality assurance is closely tracked for reporting and compliance

Incident reporting – which is consistently updated for system insights and improvement

Let’s explore each of these components in more detail.

Initial Setup

This phase starts with establishing responsibilities and access for personnel connecting to the cloud-based captioning platform. In the case of Smart Lexi, this would mean access to the iCap Cloud, where the iCap network is hosted and plays a central role in Smart Lexi’s operation.

Ai-Media’s captioning team converges with the ASR to provide training, management, and customization support. This includes providing:

  • Coordinators – who receive standard access to the cloud captioning platform and are trained on scheduling procedures.
  • Team Leaders/Managers – set up and configure each captioning instance.
  • Custom Dictionary Curators – also access the cloud captioning platform and are trained to update the custom library.

As part of the initial setup, Smart Lexi personnel integrate several supporting systems into the platform. For instance, a dedicated captioning instance application can set up one instance per captioning encoder (hardware or virtual) and apply default settings.

This settings template can be used when creating ASR captioning jobs, ensuring reliable settings duplication across different sessions. Useful settings include language, custom dictionary, access codes, and caption appearance, among others.

Remote monitoring can be set up via the cloud captioning platform’s ability to stream live captioning to smartphones, tablets and personal computers. In the case of Smart Lexi, this is called CaptionCast, which has a straightforward setup in the iCap Admin portal. Additional software features allow for easy creation, monitoring, stop, re-start, and job logging.

Administrative applications are also integrated during initial setup, using a collaborative platform to generate reports.

Custom Dictionary Management

A key component of Smart Lexi’s high performance is its Topic Models. An AI-powered feature, Topic Models greatly improve caption accuracy, particularly in the case of unusual or domain-specific words and phrases, or proper names and places.

Custom dictionaries are core to the efficient operation of a hybrid captioning platform. Ideally incorporating machine learning, these customized vocabularies consist of text-training data including:

  • Common phrases that are expected in the audio program.
  • Proper nouns which are desired to be recognized but would not be found in a standard dictionary.
  • Pronunciation clues for any new words.

The hybrid captioning platform’s accuracy can be continuously and significantly improved by careful curation of a network’s various customized libraries. For Smart Lexi, Ai-Media’s team leaders manage content teams that review all captioning transcripts, constantly updating the Topic Models custom libraries specific to a broadcaster’s programming.

The typical custom dictionary curation process consists of four main steps, which may vary according to a network’s unique needs:

  • Creation of a generic base custom dictionary and house style – during the initial onboarding process this may include the network name, program names, presenter names and recurring consistent terminology.
  • Research new terms daily – either by independent online research or by using the broadcaster’s newsroom system/materials, collating lists, adding phonetic pronunciations to terms where applicable and importing/adding terms to the custom dictionary.
  • Reviewing the previous day’s logs for discrepancies – new terms are added and outdated terms removed.
  • House styles are used to address both formatting preferences and ASR recognition errors – this is when there is high confidence that the term identified will consistently need to be replaced with another term.

The frequency of custom dictionary updates depend on the nature of the network’s programming. Some Smart Lexi user dictionaries are updated hourly.

Live Support For Scheduling And Management

Ai-media’s Smart Lexi Coordinators play a key role in maintaining its workflow. 

One of the key functions they oversee is starting and stopping Lexi instances. These can launch in several different ways. In a broadcast scenario, captioning encoders may require scheduling management by Coordinators. Since certain studio sources have fixed hours, a recurring Lexi session is scheduled. Other studio sources, however, may be subject to ad hoc hours and require daily manual scheduling.

Constant Monitoring

Coordinators monitor live captions 24/7 for dropouts and caption accuracy. Ai-Media uses the globally recognized NER model to measure caption accuracy, as it assesses live caption quality in a more objective and scientific fashion than previous models.

Incident Reporting

Collaborative tools are employed to log any on-air captioning incidents, plus add reference comments to the daily report. These incident reports are reviewed constantly, and a monthly summary is sent to the network. Archives can be stored in the cloud and downloaded easily on demand.

Bill McLaughlin, Chief Product Officer, Ai-Media.

Bill McLaughlin, Chief Product Officer, Ai-Media.

Efficiently Scaling Captions

Broadcasters, networks, and streaming content creators are all constantly creating new content. They also often have a deep archive of content that has not yet been captioned. In addition to live programming, they want to leverage as much content as possible, across multiple channels, and meeting accessibility requirements is essential. There are many more benefits to providing quality captioning as well.

A hybrid captioning system like Smart Lexi allows complex networks to build their captioning infrastructure on secure, cloud-based platforms. By adding experienced caption teams, it scales quality captions quickly, allowing users to build new content offerings to connect with viewers.