Poor dialog intelligibility is one of the biggest sources of complaints from film and television audiences, with studies suggesting that almost half of viewers prefer to watch with subtitles. But this doesn’t have to be the case! DialogCheck is an AI-powered dialog intelligibility meter, designed to provide an objective measure of speech clarity.

For an audio engineer who might have been working on a scene for hours or even days, hearing the same dialog hundreds of times, it’s a challenge to predict how a mix will sound to someone hearing it for the first time. DialogCheck combines ‘Listening Effort’ technology developed by Fraunhofer IDMT with a history view inspired by NUGEN Audio’s iconic VisLM loudness meter. The resultant plug-in, driven by dual AI neural networks and developed alongside Netflix, provides clear and intuitive data to support audio engineers in making informed decisions for their dialog mix.

What impacts dialog intelligibility?

Many of the factors that negatively impact dialog intelligibility are outside an audio engineer’s control. Speech might be in a language or dialect that the listener doesn’t understand, or might not have been enunciated clearly, the listener may be hearing impaired, or in a poor listening environment.

But there are other factors we usually do have control over; loudness of dialog, loudness of other audio (e.g. music or background noise), quality of audio signal, and the use of reverb and other effects, to name a few. If the negative impacts of these factors are mitigated as much as possible, the factors outside our control will most likely cause fewer problems.

All of these factors, both those outside the control of audio post production engineers and those within our control, fall into what Netflix's TechBlog has referred to as the Dialog Chain of Custody.

How can intelligibility be improved?

DialogCheck is a measurement tool, it does not apply any processing to the audio signal. But it does provide an objective value that acts as a signpost for where there might be room for improvement. This might mean adjusting the balance between dialog, music and effects in the mix, it might mean adjusting audio processing or using noise reduction and audio cleanup techniques to adjust the source audio. In some cases it might even mean resorting to ADR!

DialogCheck could also be integrated into a pre-production workflow in order to ensure that the source audio already reaches the desired level of clarity.

Fraunhofer LE-Meter

DialogCheck implements technology from the Listening Effort Meter developed by Fraunhofer IDMT, which uses automatic speech recognition and psychoacoustic modeling. The LE-Meter has been validated through systematic listening tests, and its metrics show a high correlation with subjective assessments of listening effort.

Typical applications

• Film audio
• Television audio
• Radio
• Podcasting
• Game audio

Video

Features

System Requirements