MusicTempo_Data_ReAnalyses_Nov2020.xlsx (158.09 kB)
Download file

Effect of Background Music Arrangement and Tempo on Foreground Speech Intelligibility: Listening experiment data and statistical analyses spreadsheets.

Download (158.09 kB)
dataset
posted on 11.05.2022, 11:38 by Philippa DemontePhilippa Demonte

Excel work book with data collected and collated from a speech-in-noise test, and associated statistical analyses. 


Background music is cited as one of the four main problems that listeners complain about with regards to foreground speech audibility in broadcast audio. Yet broadcasters, such as the BBC, only provide limited guidance to content producers regarding background music. For example, turn the background music down by 3 dB when co-present with foreground speech, and avoid lyrics and heavily percussive beats.


This quantitative, subjective listening experiment investigated whether or not background music arrangement and tempo have any effect on foreground speech intelligibility, such that additional broadcasting guidelines can be written if there are any genuine effects.


Full details of the listening experiment, results and analyses are reported in the PhD thesis by P. Demonte (2022).


KEY


5 x background music pieces 

(created with Apple Loops in Garage Band)

 - M1: legato string quartet

 - M2: solo cello; single note in a bowed, staccato style

 - M3: cello + lightly percussive instrumentation

 - M4: cello + heavily percussive instrumentation


3 x tempi

 - T1: 60 beats per minute (BPM)

 - T2: 100 bpm

 - T3: 140 bpm


Control condition: M5_T0

- purely energetic masking noise of speech-shaped noise

- for comparison against music arrangement effects. 


This speech-in-noise test used the R-SPIN speech corpus, which contains end-of-sentence target words in two semantic levels:


2 x spoken sentence semantic levels

 - HP: high predictability, e.g. "His plan meant taking a big RISK."

 - LP: low predictability, e.g. "He wants to talk about the RISK."


PID = (anonymised) participant ID #


Spreadsheet pages


* total_CWS: 

Split into several tables, including: 

 - a check of total of number of trials for each participant and each combination of the independent variables; 

 - correct word scores (1 point per correctly identified word; 1 target word per trial); 

 - mean correct word proportions; 

 - mean correct word score percentages, i.e. word recognition percentages as a proxy for quantifying speech intelligibility.


* Music_Tempo_Pred:

Word recognition percentages by participant and combination of the independent variables (music arrangement, tempo, and semantic level of sentence predictability), excluding the control conditions: M5_T0_HP and M5_T0_LP.

Statistical analyses conducted using IBM's SPSS, including:  - checks of the criteria for using 3-way Repeated Measures ANOVA;

 - simple main effects; 

 - non-parametric testing: for comparison.


Since not all of the criteria for use of 3-way RMANOVA are fulfilled, and the outcomes of the non-parametric testing were not useful, attempts were also made to transform the data (square root, squared, and arcsine transformations) and statistically re-analyse them. See spreadsheet pages:

 - SQRT_Transformed_MTP

 - ^2_Transformed_MTP

 - Arcsine_Transformed_MTP


The spreadsheet pages thereafter group the data in different ways to do 2-way and 1-way RMANOVA statistical analyses:

 - Tempo-Pred: summation across all background music pieces;

 - Music: summation across all tempi and semantic levels; 

 - Tempo: summation across all music pieces and semantic levels;

 - SentencePredictability: summation across all music pieces and tempi


The final page in this Excel work book - 'Deleted_Test' - contains data that were collected in an initial version of the listening experiment but not used towards the thesis. A quality check revealed that although all participants had completed the same total number of trials, there had been an imbalance in the number of trials per combination of independent variable. The problem was rectified in order to then conduct the listening experiment correctly. These 'Deleted_Test' data have neverthess been retained on this page in the Excel work book such that a researcher with more in-depth knowledge of other statistical methods may one day be able to analyse them for comparison.

Funding

EP/L00539/1

History