The author is CEO of NUGEN Audio.
In broadcast and radio production, few qualities are as fundamental or as deceptively complex as speech intelligibility.
It sits at the center of the listening experience, yet intelligibility is often only noticed when it fails. When dialog becomes unclear, listeners disengage, comprehension drops and the content itself loses its impact.
In radio especially, where speech is the primary carrier of information, personality and storytelling, intelligibility is not simply a technical concern; it is the core principle of audience connection.
We have seen this kind of industry-wide reckoning before. In the early 2000s, the dominant complaint from audiences was inconsistent loudness — particularly the jarring jumps between programming and commercials. Viewers were forced to ride volume controls, and broadcasters faced reputational damage and potential revenue loss as audiences switched away from uncomfortable listening experiences.
That challenge ultimately drove a global response: the development of standardized loudness measurement, most notably through ITU-R BS.1770, which finally gave engineers a way to quantify perceived loudness in a consistent, human-relevant way.
From there, regulation and practice evolved. In the U.S., legislation such as the CALM Act reinforced the need for compliance, while production and QC workflows integrated loudness normalization as standard practice. What was once subjective became measurable, and what was measurable became controllable.
Today, we may be entering a similar inflection point, only this time the issue is speech intelligibility.
Across radio, streaming and broadcast environments, complaints are shifting. Instead of “it’s too loud,” the more common listener frustration is increasingly “I can’t understand what they’re saying.”
Dialog might be masked by music and effects, compromised by poor recording conditions or degraded through complex processing chains. It might also suffer from excessive reverberation, inconsistent microphone technique, over-compression or poor performance conditions.
Unlike loudness, however, intelligibility has many more possible failure points and far less agreement on how to define or measure it.
That complexity is part of the problem. Human speech understanding is influenced not only by signal quality, but by cognition, context, language familiarity and even listener fatigue. Two audio signals with identical technical characteristics might be perceived very differently depending on content and environment.
While engineers can easily measure peaks, LUFS or dynamic range for loudness control, there has not been a universally adopted metric that captures how easily speech can be understood in real-world listening conditions.
Yet the motivation to solve this is becoming increasingly urgent. In a fragmented media environment where listener attention is fragile, poor intelligibility has direct consequences. If audiences cannot follow dialog, they are less likely to remain engaged, complete listening to a program or even return for a future episode.
For radio, where storytelling, news delivery and personality-driven formats dominate, that loss of clarity can quickly become a loss of relevance.
Encouragingly, work is underway to address this gap. New approaches to measuring “listening effort” are emerging, using machine learning models trained to estimate how difficult it is for a listener to decode speech in any given signal.
One example is the concept of a listening effort meter, such as the one found in NUGEN Audio’s DialogCheck plug-in, which analyzes short segments of audio and evaluates how confidently speech elements can be identified. Rather than relying on simple amplitude or spectral balance, these systems attempt to model perception itself — how much cognitive effort is required to understand spoken content.
Early implementations, developed across multiple technology platforms, suggest that such models can be surprisingly robust across languages and content types. That is significant because intelligibility is not bound to a single language or production style, and any useful metric must function across diverse broadcast environments. However, these tools are still evolving, and their application raises important questions.
How do you translate a series of short-term intelligibility scores into a meaningful quality threshold for a full-length program? What constitutes “acceptable” intelligibility in news versus drama versus entertainment? And, critically, how should such measurements be integrated into QC workflows without oversimplifying creative intent?
These questions mirror those faced during the loudness transition. Early loudness measurement was not immediately a production standard; it took years of refinement, testing and consensus-building before loudness became embedded in global workflows.
Speech intelligibility is likely to follow a similar path, moving gradually from experimental metrics toward standardized guidance and, eventually, operational use.
For radio broadcasters, the opportunity is clear. As tools and methodologies mature, there is potential not only to detect intelligibility issues more reliably, but to design production workflows that prevent them in the first place. That could influence microphone techniques, mixing decisions, content balancing and even editorial choices.
Ultimately, speech intelligibility is not just an audio engineering challenge; it is a listener experience issue. In radio, where the entire medium depends on the clarity of spoken word, it might also be the most important quality metric that we have not yet fully defined.
The author is founder and CEO of NUGEN Audio, a U.K- based manufacturer of professional software audio plug-ins. He founded NUGEN Audio in 2004.
Comment on this or any story to [email protected].
