Senior Staff Scientist Dolby Laboratories, Inc San Francisco, California, United States
Unguided dialog enhancement (DE) is a feature that allows a listener to increase the relative dialog level in a content item, even when only the finished mix (with no separate dialog track) is available. We introduce a new source separation technology, Spatio-Level Filtering (SLF), which we combine with dialog classification, to allow high-quality unguided DE for typical entertainment content in a stereo or higher channel count format. SLF exploits idiomatic spatial and level information and requires little lookahead, memory, computation and training data. Two subjective listening experiments indicate favorable performance.