Ch. 9 - Expressive Timing

This notebook demonstrates expressive timing profiles from real performances from https://github.com/fosfrancesco/asap-dataset.

Load libraries

library(ggplot2, quietly = TRUE)
library(tidyverse, quietly = TRUE)
library(dplyr, quietly = TRUE)

Get data from ASAP project

This gets the metadata from ASAP project (see Foscarin et al., 2020) and selects Preludes op 23.4.

d <- read.csv("https://raw.githubusercontent.com/fosfrancesco/asap-dataset/master/metadata.csv",header = TRUE,sep = ',')
df<-dplyr::filter(d,title=='Preludes_op_23_4')
df<-df[1:3,]
print(knitr::kable(head(df[,1:3],3)))
composer title folder
Rachmaninoff Preludes_op_23_4 Rachmaninoff/Preludes_op_23/4
Rachmaninoff Preludes_op_23_4 Rachmaninoff/Preludes_op_23/4
Rachmaninoff Preludes_op_23_4 Rachmaninoff/Preludes_op_23/4

Read score annotations

basedir <-'https://raw.githubusercontent.com/tuomaseerola/emr/master/'
deadpan <- read.csv(paste0(basedir,'data/midi_score_annotations.txt'),header = FALSE, sep = '\t')
print(knitr::kable(head(deadpan,3)))
V1 V2 V3
0.0 0.0 db,3/4,2
1.2 1.2 b
2.4 2.4 b 
fn <- NULL
fn[1]<-'data/ChenGuang12M_annotations.txt'
fn[2]<-'data/MorozovS09_annotations.txt'
fn[3]<-'data/WuuE07M_annotations.txt'
Performer <- c('Chen Guang','Yevgeny Morozov','Elliot Wuu')

Choose extract from all performers

D <- NULL
for (k in 1:length(fn)) {
  perf<-read.csv(paste0(basedir,fn[k]),header=F,sep='\t')
  DF<-data.frame(score=deadpan$V1,perf=perf$V1,
    annotation=deadpan$V3)
  DF <- dplyr::filter(DF,score < 30) # Limit to first 10 bars = 3*10 beats
  DF2 <- normperf(DF)                # Defined previouslys
  DF2$Performer<-Performer[k]
  D<-rbind(D,DF2)
}

options(encoding = "UTF-8")
#library(dplyr)
DF <- dplyr::filter(D,score < 30) # First 10 bars = 3*10 beats
print(knitr::kable(head(DF[,1:6],3)))
score perf annotation perf_N delta delta2
0.0 0.000000 db,3/4,2 0.000000 0.0000000 0.00000
1.2 1.916667 b 1.935339 0.7353393 735.33933
2.4 3.009115 b 3.038430 0.6384300 -96.90928

Plot expressive timing deviations

options(repr.plot.width = 12, repr.plot.height = 5)
g1 <- ggplot(DF,aes(x=perf_N,y=scoredelta_rawperf_Ndelta,colour=Performer,shape=Performer))+
  geom_line(alpha=0.85)+
  geom_point(alpha=0.85,size=2.5)+
  scale_color_grey(start = 0.30,end = 0.8)+
  geom_smooth(aes(colour = NULL,shape=NULL), method = "loess", span=0.2,se=FALSE,colour='black',linewidth=1.25)+
  scale_x_continuous(limits=c(0,30),breaks = seq(0,30,by=3),expand = c(0.02,0.002),labels=(seq(0,30,by=3)/3)+1) +
  xlab('Bar')+
  ylab('Deviation in ms')+
  theme_bw()+
  theme(legend.position=c(.85, .80))+
  theme(legend.background = element_blank()) +  # Remove overall border
  theme(legend.key = element_blank())
print(g1)

References

  • Foscarin, F., Mcleod, A., Rigaux, P., Jacquemard, F., & Sakai, M. (2020). ASAP: a dataset of aligned scores and performances for piano transcription. In International Society for Music Information Retrieval Conference (pp. 534-541).
Back to top