Aston Data Explorer

Summers, Robert

Roberts, Brian

These datasets comprise listeners’ transcriptions of sentence-length speech analogues, and the 
scores and error counts derived from them, for Experiments 1 and 2 of the article of the same title 
(Summers and Roberts, 2020; Journal of the Acoustical Society of America).  
Experiment 1: 
There is one spreadsheet for experiment 1, comprising a summary worksheet and the raw data for 
each listener. The summary worksheet contains aggregated scores (proportion keywords correct by 
tight scoring, see below) for each listener in each condition, with relevant demographic information. 
For completeness, the summary worksheet also contains the mean number of words reported by 
each listener per condition. Subsequent worksheets comprise the raw data for each listener and 
stimulus.  
The raw data comprise: (a) the stimulus presented [Column heading: Stimulus], (b) the stimulus 
sentence [Text], (c) the listener’s response [Transcription], (d) the condition number for which the 
stimulus was presented [Condition], (e) the number of times the listener heard the stimulus (always 
once in this experiment), (f) the number of keywords in the stimulus, (g) the loose score (number of 
keywords reported correctly for which the stem of the word is correct – e.g., “type”, “types”, and 
“typed” would all be marked correct for keyword “typing”; the loose score was not analysed but is 
included for completeness), and (h) the tight score (only exactly reported keywords are marked as 
correct; homonyms are accepted). The mean scores for each condition – (i) loose and (j) tight – are 
computed by dividing the number of correct keywords reported for all sentences in the condition (6 
sentences/condition) by the total number of keywords. Additionally, column (l) [Number of words 
reported] computes the number of words transcribed by the listener regardless of their correctness 
and (m) computes the mean number transcribed for each condition. 
Experiment 2:  
There are two spreadsheets for experiment 2, one for the main experiment and one for the follow-
up experiment.  
The spreadsheet for the main experiment comprises three summary worksheets and the raw data 
for each listener. The SummaryTarget worksheet contains aggregated scores (proportion keywords 
correct by tight scoring) for each listener in each condition (C1-C7) and also collapsed across target-
to-masker ratio for the conditions involving interferers (C1-C3 vs C4-C6), with relevant demographic 
information. The SummaryInterferer worksheet contains aggregated scores (proportion of keywords 
reported that are exclusively in the interferer by tight scoring). Finally, the SummaryError worksheet 
contains the count of two types of error – intrusion errors and other errors (for both loose and tight 
scoring). Intrusion errors are the number of keywords reported that were exclusively in the 
interferer(*). Other errors are the number of words reported that would usually be classified as 
keywords but which were not present in either the interferer or the target. 
(*) Owing to small differences in the number of keywords per asynchrony condition (54 or 55), note 
that the correspondence between the proportion of keywords reported that are exclusively in the 
interferer (SummaryInterferer) and the intrusion error count (SummaryError) is close but not exact. 
Subsequent worksheets comprise the raw data for each listener and stimulus. The raw data 
comprise: (a) the stimulus presented [Column heading: Stimulus], (b) the listener’s response 
[Transcription], (c) The text of the target sentence [TargetText], (d) the text of the interferer 
sentence [InterfererText] where applicable, (e) the condition number for which the stimulus was 
presented [Condition], (f) the number of times the listener heard the stimulus (always once in this 
experiment), (g) the number of keywords in the target, (h) the loose score (number of target 
keywords reported correctly for which the stem of the word is correct – e.g., “type”, “types”, and 
“typed” would all be marked correct for keyword “typing”), (i) the tight score (only exactly reported 
target keywords are marked as correct; homonyms are accepted), (j) the number of keywords in the 
interferer, (k) the number of words reported that were exclusively in the interferer according to 
loose scoring, (l) the number of words reported that were exclusively in the interferer according to 
tight scoring, (m) the number of intrusion errors based on loose scoring [ErrorIntrusionLoose], (n) 
the number of words reported that were not in either the interferer or the target based on loose 
scoring but which would otherwise be classified as keywords [ErrorOtherLoose], (o) the number of 
intrusion errors based on tight scoring [ErrorIntrusionTight], (p) the number of words reported that 
were not in either the interferer or the target based on tight scoring but which would otherwise be 
classified as keywords [ErrorOtherTight], (r) the mean tight scores for each condition are computed 
by dividing the number of correct keywords reported for all sentences in the condition (6 
sentences/condition) by the total number of keywords, (t) the mean interferer scores for each 
condition (tight scoring) are computed by dividing the number of keywords reported that appeared 
exclusively in the interferer for all interfering sentences in the condition (6 sentences/condition) by 
the total number of interferer keywords. The final four columns – (v) [SumErrorIntrusionLoose], (w) 
[SumErrorOtherLoose], (y) [SumErrorIntrusionTight], and (z) [SumErrorOtherTight] – are the total 
number of those errors for each condition. 
The spreadsheet for the follow-up experiment comprises a summary worksheet and the raw data for 
each listener. The summary worksheet contains aggregated scores (proportion keywords correct by 
tight scoring, see below) for each listener in the two formant-asynchrony conditions – C1-C3 (0 ms) 
vs C4-C6 (150 ms) –  and the mean number of words reported by each listener per asynchrony 
condition. Subsequent worksheets comprise the raw data for each listener and stimulus.  
The raw data comprise: (a) the stimulus presented [Column heading: Stimulus], (b) the stimulus 
sentence [Text], (c) the listener’s response [Transcription], (d) the condition number for which the 
stimulus was presented [Condition] (1 to 3 correspond to 0-ms formant asynchrony and 4-6 
correspond to 150-ms formant asynchrony), (e) the number of times the listener heard the stimulus 
(always once in this experiment), (f) the number of keywords in the stimulus, (g) the loose score 
(number of keywords reported correctly for which the stem of the word is correct – e.g., “type”, 
“types”, and “typed” would all be marked correct for keyword “typing”; the loose score was not 
analysed but is included for completeness), and (h) the tight score (only exactly reported keywords 
are marked as correct; homonyms are accepted). The mean scores for each asynchrony condition – 
(i) loose and (j) tight – are computed by dividing the number of correct keywords reported for all 
sentences in the condition (6 sentences/condition, corresponding to 18 sentences for each formant 
asynchrony) by the total number of keywords. Additionally, column (l) [Number of words reported] 
computes the number of words transcribed by the listener regardless of their correctness, and (m) 
computes the mean number transcribed for each asynchrony condition (C1-C3 vs C4-C6). 
 


English

Informational masking of speech by acoustically similar intelligible and unintelligible interferers

Gaur, Chandra

Ferreira, Filipe

Gordienko, Vladimir

Ribeiro, Vitor

Szabo, Aron

Doran, Nick

Experimental comparison of fiber optic parametric, Raman and Erbium amplifiers for burst traffic for extended reach PON

Neskorniuk, Vladislav

Freire, Pedro J.

Napoli, Antonio

Spinnler, Bernhard

Schairer, Wolfgang

Prilepsky, Jaroslaw E.

Costa, Nelson

Turitsyn, Sergei K.

The example codes and numerical traces supplemented for article "Data augmentation for the supervised learning of nonlinearity compensation algorithms in optical communications"

Gulamhussein, Aiman

Uddin, Romez

Tighe, Brian

Poyner, David

Rothnie, Alice

Data underpinning "A comparison of SMA (styrene maleic acid) and DIBMA (di-isobutylene maleic acid) for membrane protein purification".

Nguyen, Tu

Coupled Transceivers-Fiber Nonlinearity Compensation Based on Machine Learning for Probabilistic Shaping System

Yang, Yang

A Fundamental Study on the CO2 Gasification of Char Derived from Miscanthus at Various Processing Conditions

Lamb, Jacob R.

Szabó, Áron

Data underpinning 'Phase-sensitive amplification of 11 WDM channels across bandwidth of 8 nm in a fibre optic parametric amplifier'

Krzczanowicz, Lukasz

Iqbal, Md Asif

Phillips, Ian

Harper, Paul

Forysiak, Wladek

High-capacity multi-span transmission performance characterization of broadband discrete Raman amplifier [data underpinning article]

Steam gasification of Miscanthus derived char: the reaction kinetics and reactivity with correlation to the composition and microstructure

Fiber Optic Parametric Amplifier for High Capacity Busty Networks

Aston University

Not a member yet