Crystal Clear: Boosting Speech Intelligibility in Media
Project Objectives
Champions: BBC, Channel 4, Fraunhofer IDMT, ITV, Sky, IET, AES, RNID, University of Salford
Participants: Shure
Crystal Clear aimed to establish a consistent, practical way to measure and improve dialogue intelligibility across broadcast and streaming.
The project defined a unified metric for audibility by combining machine model based measurement tools with structured human listening assessments, creating a clear and reliable baseline for how dialogue intelligibility was evaluated.
It then explored how this metric could be applied within production and distribution workflows to support more consistent and measurable improvements in dialogue quality.
Innovation and Collaboration
Across the industry, dialogue intelligibility was already being measured using a range of tools, but these often produced inconsistent or non-comparable results, making it difficult to apply them reliably or consistently. There was a clear need to bring greater alignment to how audibility was assessed by combining assessment with machine models and structured human listening assessments to establish a shared point of reference. Aligning technical outputs with how content was judged by listeners enabled the development of a more consistent and practical framework for assessing dialogue clarity across different environments and use cases. This also created a foundation for applying these approaches within production and distribution workflows, testing how they performed in different technical and operational environments.
This required close collaboration between broadcasters, technology providers, and research partners. Broadcasters contributed real content and workflows, technology providers brought measurement and processing tools, and research partners supported the design and validation of listening assessments. Shure also supported the research by providing equipment as incentives for participants in the listening study, helping enable the large-scale audience research that underpinned the project. Working together in this way ensured that approaches were tested end-to-end, grounded in real-world conditions and capable of wider adoption across the industry.
Proof of Concept Demonstrated at IBC:
The Crystal Clear proof of concept demonstrated how dialogue intelligibility measurement could be integrated directly into a real broadcaster quality-control workflow, enabling potential issues to be automatically identified and presented for targeted human review.
A critical part of delivering the proof of concept was the research that underpinned it. 300 participants - 150 normal-hearing and 150 hard-of-hearing - assessed 60 clips from real broadcast content using a structured listening-effort scale. The same clips were analysed using a wide range of machine-based measures, creating a substantial body of human listening data against which objective models could be compared and validated.
This provided evidence of which automated approaches most closely reflected real audience experience and enabled the project to investigate practical measurement ranges and thresholds for identifying content that may require further review. While the research found that a single definitive threshold could not yet be applied consistently across all content, it demonstrated that low objective scores can be used to reliably flag potentially problematic dialogue for human assessment.
The resulting approach was implemented within Channel 4’s Media Services environment. Incoming programme content was automatically analysed after ingest, with timecoded measurement results associated with the programme asset in Channel 4’s BRAHMA AI asset management environment. Reviewers could then focus on flagged sections rather than listening to the entire programme, demonstrating how the research could translate into a practical, scalable human-in-the-loop QC workflow.
Project PoCs:
View the 2027 Proof of Concept assets the team of Crystal Clear: Boosting Speech Intelligibility in Media presented at IBC2026.
Key Learnings:
The research demonstrated that dialogue intelligibility and listening effort can be assessed using automated measurement, but that no single machine-based measure provides a complete representation of audience experience. Across the models tested, several showed meaningful correlation with human listening assessments, with speech-to-text based measures producing the strongest correlations within the Crystal Clear corpus.
The study also highlighted an important difference in audience experience. Hard-of-hearing participants rated the same content around 1.5 points more difficult to listen to than normal-hearing participants, reinforcing the importance of including different hearing experiences when developing approaches to dialogue quality assessment.
The project found that objective measurements are particularly valuable as a flagging mechanism rather than as a replacement for human judgement. Low measurement values can identify sections of content that are more likely to require attention, allowing automated analysis to direct reviewers towards potential issues. However, a single universal threshold could not yet be established across all programme genres and content types, demonstrating the need for further research and refinement.
Testing within an operational workflow also provided practical learnings about how this could work at scale. Reviewers benefited from additional context around flagged segments, while filtering and configurable thresholds will be important to ensure that systems surface a manageable number of relevant issues without creating unnecessary operational overhead.
Industry Impact:
Crystal Clear has created a significant evidence base for moving dialogue intelligibility from a largely subjective concern towards a measurable and operational part of media quality control.
The combination of listening assessments from 300 people, including both normal-hearing and hard-of-hearing participants, with objective measurements across the same real-world broadcast material provides a dataset against which machine-based approaches can be evaluated according to how closely they reflect actual audience experience. This creates a foundation for further development of practical thresholds, measurement approaches and QC methodologies.
For broadcasters and content producers, this opens the possibility of identifying dialogue issues earlier and at scale, using automated analysis to focus human attention where it is most valuable while continuing to support creative and editorial judgement.
Most importantly, improving dialogue intelligibility has the potential to make television and streamed content more accessible for everyone. Clearer, easier-to-understand dialogue benefits all audiences, while being particularly important for people who are deaf or hard of hearing. By incorporating the listening experiences of hard-of-hearing audiences directly into the research, Crystal Clear provides a foundation for future measurement approaches and quality standards that better reflect the needs of a wider range of viewers.
More broadly, the research and operational proof of concept provide a foundation for greater industry alignment around how dialogue intelligibility is measured and managed, with the potential to inform common thresholds, delivery specifications and interoperable measurement approaches across the industry.
){kind=logo}
)
)
)
)
){kind=logo}
)
)