Automation in Games

Efficiency in a creative industry

A quiet revolution is happening in audio recording technology as platforms move to the cloud, offer productivity improvements and introduce some of the benefits we have taken for granted in text localization for years. At Lionbridge Games, we are helping to drive that process with our cloud production platform built specifically for games voice audio production.

Automation can improve productivity, reduce human error and enhance the user’s experience by simplifying their tasks while removing some of the boring bits. But automation can also hinder the creative process—how many times has the predictive text feature on your phone tried to ruin your most creative messages with a standardized, rote response?   

With this experience, it's not surprising that suggesting the intrusion of automation into the creative domain can cause some serious consternation. Creativity is about breaking rules, having the freedom to implement an expert decision and (sometimes quite literally) not stick to the script. How do we manage these competing demands?  

Connection: Things and People 

While connecting things and people might not feel like automation (which in a creative industry can be a good thing!), it achieves the same end: eliminating human processes that add little value. A lot of audio production workflows in the past involved moving content and data across global studios and between people in the same team and different toolsets.   

By bringing tools and teams under the umbrella of a single production platform, we eliminate the transition work: The output of a step becomes the input for another. So, in Lionbridge Games Cloud for Audio, the content that is ingested for script management by the central team is the input for localization by all teams and is used for automated assets checks. These assets are used to generate recording sessions, which are used in batch processing and for QA, among other uses.   

And connecting workflow steps also connects team members that might previously never have opened the same software: the linguist performing script adaptation, the voice director in the studio and the post-production sound engineer or the LQA tester. In our games cloud solution, they are right there, in real time, in the same toolset and can communicate through comments and issues appended to any line. This removes the need to go through yet another toolset for communication and creates a collaborative space that enhances the creative process.   

Connecting things and people is an unintrusive way to achieve automation goals while reinforcing creative collaboration. And you don’t even need to use the A-word!  

Automation Designed for Integration with Expert Processes

Ask a new sound engineer, six weeks out of college, how to post produce game audio files and they will likely give you a nice, clear, sequential list of steps—an automation dream! Ask the same engineer five years later and they’re more likely to wonder aloud, “What does that question even mean?”   

It’s a phenomenon first encountered by software teams working in expert systems in the 70s. Experts don’t follow paths, they make them: constantly, often unconsciously, adapting their processes to the specifics of the problem at hand.  If experts are hard to automate for, creative experts...ugh!  

So, what can you do? Firstly, any automation needs to recognize that it is not the expert here. Automation kicks in when the expert decides, and the expert determines the parameters. For example, Lionbridge Games Cloud for Audio perfectly understands the games’ standard recording duration constraints, but if the expert wants “Match Source Duration +/- 27%,” the software will dutifully oblige, and alert users whenever the target audio fails to meet that constraint.   

Secondly, experts will frequently adapt and tweak processes to every line, so Lionbridge Games Cloud automations and batch processes can be configured individually for each line of audio, and mini-batch processes can be quickly run on any selection of lines while overriding the configuration. Or perhaps, the best solution is currently undetermined because information is unavailable. Voice Directors using Lionbridge Games Cloud in the studio can insert additional alternative takes as new lines in a single click.   

Giving creative teams quick-to-configure and quick-to-deploy automations in smaller chunks helps to integrate the time-saving steps as building blocks within the process that the expert is typically building on the fly. It ensures that the experts stay in control, and it allows them to break the rules as needed. The Lionbridge Games Cloud’s automated checks will alert relevant teams, including QA, if and when rules have been broken. And the built-in communications enable everyone to understand the reasons, and to confirm that those reasons are justified.  

It ensures that the experts stay in control, and it allows them to break the rules as needed.

Augment and Assist Experts  

This human-centered approach to automation recognizes that sometimes creative experts are just plain better at a task than any available automation. However, there may still be opportunities to improve user performance.   

For example, script vs audio mismatches are the most common of all audio bugs. QA and LQA teams know there are mismatches in the content—they just have to find them. It’s a needle-in-the-haystack problem that typically requires listening to each audio file multiple times. Our games cloud uses speech-to-text (STT) to extract the content of audio files and compares the resulting string against the script to see if they match.   

Although STT performs impressively for film and video conferencing, it can struggle with the unusual sentences, vocabulary and generally short lines found in games. This can generate a high number of false negatives. We made significant gains in accuracy with our enhanced STT+ technology for detecting mismatches and, with some variation between languages, about 85 percent of lines can be definitively flagged “Good to go!” The remaining 15 percent of the haystack is handed back to the humans.   

However, now armed with a visual red/blue comparison of the audio and script strings, the user can visually scan the content and listen only to those parts of the audio flagged as a potentially mismatch. Finding all the needles has been simplified by combining automation with human expertise, and by presenting information in the form that is easiest for the user to process.  

We adopt a similar approach to detect breach of timing constraints: an automated check identifies potential issues, and visual cues simplify the human task of confirming whether a fix is required. For sound sync recordings, where pauses in speech need to align between source and target audio, the source and target waveforms are overlaid, and the pauses (silences) are marked visually. Users can see how closely the pauses align and make an expert decision on whether to they align sufficiently for the given context.  

Balance for the Better

As game development teams have always known, appropriate automation has a role in creative industries, and that includes game voice production. And ‘appropriate’ is the operative word: Automation can’t replace the human touch. It should be applied to take over mundane, repetitive tasks and only augment creative tasks that humans do better.    

Rather than the autocorrect feature on your phone that makes changes of its own accord, we need more tools like the spell check in your document editor that highlights possible errors. Tools that leave the expert in control, have your back and give you the freedom to break rules and get creative.  

linkedin sharing button
  • #audio
  • #blog_posts
Nicolas Underwood