In Cdevenish/hardRain: Automated detection of rain in audio recordings

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  eval = FALSE
)

##### NOT RUN ######

This vignette will download an additional data set for training and testing. It will be stored within the root directory of the hardRain package. Run the R chunks to download and use this extra testing data. The data is hosted on an OSF project: https://osf.io/qpev7/

library(hardRain)
library(ggplot2)
library(tidyr)

1. Obtain wav filenames

NOTE: this will download 200 15s wav files (252 MB) to a new directory.

Option 1. Download using osfr library. This will create a directory named 'extra_data' in the hardRain root folder in your R library location. You must have write permission to this folder.

library(osfr)

hr_path <- system.file(package = "hardRain")
extra_data_dir <- file.path(hr_path, "extra_data")

if(!dir.exists(extra_data_dir)){

  # get osf project
  hardRain <- osf_retrieve_node("https://osf.io/qpev7/")

  # list files on hardRain osf project
  ls_files <- osf_ls_files(x = hardRain)
  ls_files

  # get id for extra_data folder
  id <- ls_files$id[ls_files$name == "extra_data"]

  # get directory 
  extra_data_osf <- osf_retrieve_file(id)

  # download data to package root in a new 'extra_data' directory
  hardRain_osf2 <- osf_download(x = extra_data_osf,
                                path = hr_path,
                                recurse = TRUE,
                                conflicts= "skip",
                                progress = TRUE)

  } else warning("Directory already exists - check contents!")

Option 2. Download the complete data set from here as a zip and store in your desired location and set this variable to the 'extra_data' folder location

# extra_data_dir <- "PATH/TO/extra_data"

Rain and test data:
- rain data: 100 wav files known to be hard rain
- test data: 100 wav files - 50 rain, 50 non-rain (clear)

Get the filenames of the training and test data:

train  <- list.files(path = file.path(extra_data_dir, "rain"), pattern = "\\.wav$", full.names = TRUE)
test <- list.files(path = file.path(extra_data_dir, "test"), pattern = "\\.wav$", full.names = TRUE)

2. Calculate thresholds on training data

Calculate the threshold using default settings - for two frequency bands

tr <- getThreshold(train)
tr

Inspect the Power Spectrum Density (psd) and Signal to Noise values (s2n) for rain and non-rain files at specific frequency bands

met <- getMetrics(test)
head(met)

# bind into a data frame with the filenames
psd_s2n <- data.frame(filename = rownames(met), met)
rownames(psd_s2n) <- NULL

summary(psd_s2n)

# add test status
psd_s2n$rain <- ifelse(grepl("Clear", psd_s2n$filename), FALSE, TRUE)
head(psd_s2n)

# reorganise
metrics <- tidyr::pivot_longer(psd_s2n, c(band.1.psd:band.2.s2n),
                               names_to = c("band", "metric"),
                               names_pattern = "(band\\.1|band\\.2)\\.(.*)",
                               values_to = "value")

head(metrics)

Draw a boxplot to see separation between rain and non-rain for each metric

ggplot(metrics, aes(x = band, y = value, fill = rain))+
  geom_boxplot()+
  facet_wrap(~metric, scales = "free_y")+
  ylab("Metric value")+
  xlab("Band")

3 Classify the test files

Classify the test files using the thresholds obtained above

res <- classifyRain(test, thresh.vals = tr, threshold = c("min", "Q2"), ID = !grepl("Clear", basename(test)))
head(res)

How many files identified as rain/non-rain?

table(res[c("ID", "value", "threshold")])

Cdevenish/hardRain documentation built on Jan. 10, 2025, 3:39 a.m.