Downward Dynamic Range Expansion versus Ducking: 2-AC Test Data
Data collected towards the PhD thesis by P. Demonte (2022), which evaluated the effectiveness of several different object-based approaches to audio towards improving speech intelligibility, primarily for application to broadcast audio.
Part one of the second investigation towards this thesis involved a subjective, quantitative speech-in-noise test, which determined that when foreground dialogue and background music are co-present in a sound scene, the application of even a small amount of downward dynamic range expansion (DDRE) just to the background music can significantly improve the intelligibility of the speech compared to when DDRE is not applied.
The application of DDRE attenuates sounds below a pre-designated gain threshold; it makes quiet sounds quieter. When applied to background music, this can create more space for foreground dialogue to be glimpsed (acoustically heard), but without losing the narrative intent of the music, which is particularly important for documentary and drama programmes.
However, the greater the amount of DDRE applied, as controlled by the gain threshold, DDRE ratio, attack and release time attributes, the greater the likelihood of introducing unwanted audible artefacts.
As an end-user study, part two of this second investigation involved a Two Alternative Choice (2-AC) Test in an attempt to answer the research questions:
Does the application of downward dynamic range expansion on background music have an effect on:
1) the perceived sound quality of the background music?
2) preference of the overall speech-music sound mix?
For each trial of the 2-AC Test, participants listened to pairs of audio stimuli - A and B, with both containing foreground speech and background music - and for each attribute (sound quality and preference) selected one response: either A (best), (both A and B sound the) same, or B (best).
The background music stimuli in each trial either had one of four different levels of DDRE applied, or one of four comparable different levels of ducking (linear attenuation) applied, or no audio engineering manipulation applied.
See the 'Read Me' page of the Excel spreadsheet for a complete overview of the research questions, how the 2-AC listening experiment was conducted, how the sound quality and preference data were analysed using two different methods, and the relevant references for these methodologies. The 'Read Me' page also provides an overview of the data presented on each worksheet page.
For further information, contact:
See also the other entries of Philippa Demonte's Salford Figshare profile.