Service-Side Dialogue Clarity for Audio Streaming with Backward-Compatible Delivery and Adaptive Rendering

Briand M. Service-Side Dialogue Clarity for Audio Streaming with Backward-Compatible Delivery and Adaptive Rendering. In: AES Show 2026 Nashville. Audio Engineering Society; 2026.pp. 23:1-12. Available from: https://aes.org/publications/elibrary-page/?id=23457

@inproceedings{Briand2026_23457,
  author = {Briand, Manuel},
  title = {{Service-Side Dialogue Clarity for Audio Streaming with Backward-Compatible Delivery and Adaptive Rendering}},
  booktitle = {AES Show 2026 Nashville},
  pages = {23:1-12},
  url = {https://aes.org/publications/elibrary-page/?id=23457},
  year = {2026},
  month = oct,
  publisher = {Audio Engineering Society},
  address = {Gaylord Opryland Resort}
}

TY  - CPAPER
TI  - Service-Side Dialogue Clarity for Audio Streaming with Backward-Compatible Delivery and Adaptive Rendering
AU  - Briand, Manuel
T2  - AES Show 2026 Nashville
SP  - 1
EP  - 12
M1  - Paper 23
PY  - 2026
DA  - 2026/10/30
UR  - https://aes.org/publications/elibrary-page/?id=23457
PB  - Audio Engineering Society
CY  - Gaylord Opryland Resort
LA  - en
AB  - Disney+ ships Cinematic Audio authored for calibrated playback, e.g., near-field and/or theatrical, but most listening happens on everyday devices in uncontrolled environments. Subtitle use is a common workaround when dialogue is hard to follow [19]. Elevated music-to-dialogue ratios, downmix compression, and listener variability (device, environment, and hearing ability) further reduce intelligibility across the general population [18]. This paper presents the Dialogue Clarity Engine (DCE), a codec-agnostic server-side pipeline combining neural source separation, targeted loudness enhancement, and professional Dynamic Range Control (DRC) (CODA Edge, Skywalker Sound). An objective evaluation on fifteen Cinematic masters compares four processing-order variants (Types AD) under CODA Edge DRC at -24 LKFS, with Extended Short-Time Objective Intelligibility (eSTOI) scored at loudness-matched source integrated level (~-27 LKFS). Production defaults to Type D with CODA Edge at -24 LKFS, with an optional -16 LKFS Accessible boost. Type D achieves the highest mean eSTOI improvement (+10.5 points) among Types AD. Enhanced Dialogue is packaged for client-side crossfade in the PCM domain after existing AAC-LC / E-AC-3 decoders (Section 6). Section 5.4 reports a non-blind Skywalker Sound expert listen at -24 LKFS; formal perceptual tests are planned (Section 8.2).
ER  -