| Title: | Coding Sound Files for Use with NDL |
|---|---|
| Description: | Make acoustic cues to use with the R packages 'ndl' or 'ndl2'. The package implements functions used in the PLoS ONE paper: Denis Arnold, Fabian Tomaschek, Konstantin Sering, Florence Lopez, and R. Harald Baayen (2017). Words from spontaneous conversational speech can be recognized with human-like accuracy by an error-driven learning algorithm that discriminates between meanings straight from smart acoustic features, bypassing the phoneme as recognition unit. PLoS ONE 12(4):e0174623 <doi:10.1371/journal.pone.0174623> More details can be found in the paper and the supplement. 'ndl' is available on CRAN. 'ndl2' is available by request from <[email protected]>. |
| Authors: | Denis Arnold [aut, dtc, cre], Elnaz Shafaei Bajestan [ctb] |
| Maintainer: | Denis Arnold <[email protected]> |
| License: | GPL (>= 2) |
| Version: | 1.0.2 |
| Built: | 2024-11-20 03:19:03 UTC |
| Source: | https://github.com/denis-arnold/acousticndlcoder |
Package to make acoustic cues to use with ndl or ndl2.
The packages main function is makeCues. readTextGridFast,
readTextGridRobust, readESPSAnnotation and
readWavesurfer are helper functions that read the corresponding annotation
files and return a data.frame. CorpusCoder codes a whole corpus given a vector
with the path to and names of wave files and a vector for the annotation files.
word_classification_data provides data from Arnold et al 2017
https://doi.org/10.1371/journal.pone.0174623
Denis Arnold
Reference to to paper in accepted form.
## Not run: # assuming the corpus contains wave files and praat textgrids setwd(~/Data/MyCorpus) # assuming everything is in one place #assuming you have one wav for each annotation Waves=list.files(pattern="*.wav",recursive=T) Annotations=list.files(pattern="*.TextGrids",recursive=T) # see above # Lets assume the annotation is in UTF-8 and you want everything from a tier called words # Lets assume tha you want to dismiss everything in <|> # Lets assume that have 4 cores available # Lets assume that you want the defaut settings for the parameters Data=CorpusCoderCorpusCoder(Waves, Annotations, AnnotationType = "TextGrid", TierName = "words", Dismiss = "<|>", Encoding, Fast = F, Cores = 4, IntensitySteps = 5, Smooth = 800) ## End(Not run)## Not run: # assuming the corpus contains wave files and praat textgrids setwd(~/Data/MyCorpus) # assuming everything is in one place #assuming you have one wav for each annotation Waves=list.files(pattern="*.wav",recursive=T) Annotations=list.files(pattern="*.TextGrids",recursive=T) # see above # Lets assume the annotation is in UTF-8 and you want everything from a tier called words # Lets assume tha you want to dismiss everything in <|> # Lets assume that have 4 cores available # Lets assume that you want the defaut settings for the parameters Data=CorpusCoderCorpusCoder(Waves, Annotations, AnnotationType = "TextGrid", TierName = "words", Dismiss = "<|>", Encoding, Fast = F, Cores = 4, IntensitySteps = 5, Smooth = 800) ## End(Not run)
Helper function for makeCues
CODE(SPEC, num)CODE(SPEC, num)
SPEC |
Spectrum representation made in makeCues() |
num |
Number of the part |
A string containing the coding. Each band is seperated by "_".
Denis Arnold
Codes a corpus for use with NDL with vector of wavefile names and a vector of TextGrid names provided
CorpusCoder(Waves, Annotations, AnnotationType = c("TextGrid", "ESPS"), TierName = NULL, Dismiss = NULL, Encoding, Fast = F, Cores = 1, IntensitySteps, Smooth)CorpusCoder(Waves, Annotations, AnnotationType = c("TextGrid", "ESPS"), TierName = NULL, Dismiss = NULL, Encoding, Fast = F, Cores = 1, IntensitySteps, Smooth)
Waves |
Vector with names (and full path to if not in wd) of the wave files. |
Annotations |
Vector with names (and full path to if not in wd) of the TextGrid files. |
AnnotationType |
Type of annotation files. Suported formats are praat TextGrids (set to "TextGrid") and ESPS/Wavesurfer (set to "ESPS") files. |
TierName |
Name of the tier in the TextGrid to be used. |
Dismiss |
Regular expression for Outcomes that should be removed. Uses grep. E.g. "<|>" would remove <noise>,<xxx>, etc. Default is NULL. |
Encoding |
Encoding of the annotation file. It is assumed, that all annotation files have the same encoding. |
Fast |
Switches between a fast and a robust TextGrid parser. For Fast no "\n" or "\t" may be in the transcription. Default is FALSE. |
Cores |
Number of cores that the function may use. Default is 1. |
IntensitySteps |
Number of steps that the intensity gets compressed to. Default is 5 |
Smooth |
A parameter for using the kernel smooth function provied by the package zoo. |
A data.frame with $Cues and $Outcomes for use with ndl or ndl2.
Denis Arnold
## Not run: # assuming the corpus contains wave files and praat textgrids setwd(~/Data/MyCorpus) # assuming everything is in one place #assuming you have one wav for each annotation Waves=list.files(pattern="*.wav",recursive=T) Annotations=list.files(pattern="*.TextGrids",recursive=T) # see above # Lets assume the annotation is in UTF-8 and you want everything from a tier called words # Lets assume tha you want to dismiss everything in <|> # Lets assume that have 4 cores available # Lets assume that you want the defaut settings for the parameters Data=CorpusCoderCorpusCoder(Waves, Annotations, AnnotationType = "TextGrid", TierName = "words", Dismiss = "<|>", Encoding, Fast = F, Cores = 4, IntensitySteps = 5, Smooth = 800) ## End(Not run)## Not run: # assuming the corpus contains wave files and praat textgrids setwd(~/Data/MyCorpus) # assuming everything is in one place #assuming you have one wav for each annotation Waves=list.files(pattern="*.wav",recursive=T) Annotations=list.files(pattern="*.TextGrids",recursive=T) # see above # Lets assume the annotation is in UTF-8 and you want everything from a tier called words # Lets assume tha you want to dismiss everything in <|> # Lets assume that have 4 cores available # Lets assume that you want the defaut settings for the parameters Data=CorpusCoderCorpusCoder(Waves, Annotations, AnnotationType = "TextGrid", TierName = "words", Dismiss = "<|>", Encoding, Fast = F, Cores = 4, IntensitySteps = 5, Smooth = 800) ## End(Not run)
Helper function for makeCues that splits the signal based on the envelope of the signal
getBoundary(Wave, smooth = 800)getBoundary(Wave, smooth = 800)
Wave |
A Wave object (see tuneR) |
smooth |
A parameter for using the kernel smooth function provied by the package zoo. |
A vector with the sample numbers of the boundaries.
Denis Arnold
## Not run: library(tuneR) Wave=readWave("MyWaveFile.wav") Boundaries=getBoundary(Wave,800) ## End(Not run)## Not run: library(tuneR) Wave=readWave("MyWaveFile.wav") Boundaries=getBoundary(Wave,800) ## End(Not run)
Creates a string with the cues for each frequency band and segment seperated by "_"
makeCues(WAVE, IntensitySteps = 5, Smooth = 800)makeCues(WAVE, IntensitySteps = 5, Smooth = 800)
WAVE |
A Wave object (see tuneR). Currently it is implemented for use with 16kHz sampling rate. |
IntensitySteps |
Number of steps that the intensity gets compressed to. Default is 5. |
Smooth |
A parameter for using the kernel smooth function provied by the package zoo. |
A string containing the coding. Each band and part is seperated by "_"
Denis Arnold
## Not run: library(tuneR) library(seewave) Wave=readWave("MyWaveFile.wav") if([email protected]!=16000){ Wave=resamp(Wave,[email protected],g=16000,output="Wave") } Cues=makeCues(Wave,IntensitySteps=5,Smooth=800) ## End(Not run)## Not run: library(tuneR) library(seewave) Wave=readWave("MyWaveFile.wav") if(Wave@samp.rate!=16000){ Wave=resamp(Wave,f=Wave@samp.rate,g=16000,output="Wave") } Cues=makeCues(Wave,IntensitySteps=5,Smooth=800) ## End(Not run)
Reads a ESPS/Old Wavesurfer style annotation file and returns a data.frame with times and lables
readESPSAnnotation(File, Encoding)readESPSAnnotation(File, Encoding)
File |
Name (with full path, if not in wd) of the annotation file |
Encoding |
Encoding of the annotation file. Typically encodings are "ACSII","UTF-8" or "UTF-16" |
A data.frame with $Output for the lable $start and $end time of the lable.
Denis Arnold
## Not run: # Assume that NameOfAnnotation is encoded in "UTF-8" Data=readESPSAnnotation("NameOfTextGrid","UTF-8") ## End(Not run)## Not run: # Assume that NameOfAnnotation is encoded in "UTF-8" Data=readESPSAnnotation("NameOfTextGrid","UTF-8") ## End(Not run)
Reads a TextGrid made with praat and returns a list with a vector of all tier names and a data.frame for each tier.
readTextGridFast(File, Encoding)readTextGridFast(File, Encoding)
File |
Name (with full path, if not in wd) of the TextGrid |
Encoding |
Encoding of the TextGrid. Typically encodings are "ACSII","UTF-8" or "UTF-16" |
This method has sometimes problems with certain sequences like "\n" in the annotation file. If the method fails, try readTextGridRobust()
A list containing a vectors with the names and data.frames for each tier in the TextGrid.
Denis Arnold
## Not run: # Assume that NameOfTextGrid is encoded in "UTF-8" Data=readTextGridFast("NameOfTextGrid","UTF-8") ## End(Not run)## Not run: # Assume that NameOfTextGrid is encoded in "UTF-8" Data=readTextGridFast("NameOfTextGrid","UTF-8") ## End(Not run)
Reads a TextGrid made with praat and returns a list with a vector of all tier names and a data.frame for each tier
readTextGridRobust(File, Encoding)readTextGridRobust(File, Encoding)
File |
Name (with full path, if not in wd) of the TextGrid |
Encoding |
Encoding of the TextGrid. Typically encodings are "ACSII","UTF-8" or "UTF-16" |
A list containing a vectors with the names and data.frames for each tier in the TextGrid.
Denis Arnold
## Not run: # Assume that NameOfTextGrid is encoded in "UTF-8" Data=readTextGridRobust("NameOfTextGrid","UTF-8") ## End(Not run)## Not run: # Assume that NameOfTextGrid is encoded in "UTF-8" Data=readTextGridRobust("NameOfTextGrid","UTF-8") ## End(Not run)
Reads a New Wavesurfer style annotation file and returns a data.frame with times and lables
readWavesurfer(File, Encoding)readWavesurfer(File, Encoding)
File |
Name (with full path, if not in wd) of the annotation file |
Encoding |
Encoding of the annotation file. Typically encodings are "ACSII","UTF-8" or "UTF-16" |
A data.frame with $Output for the lable $start and $end time of the lable.
Denis Arnold
## Not run: # Assume that NameOfAnnotation is encoded in "UTF-8" Data=readWavesurfer("NameOfTextGrid","UTF-8") ## End(Not run)## Not run: # Assume that NameOfAnnotation is encoded in "UTF-8" Data=readWavesurfer("NameOfTextGrid","UTF-8") ## End(Not run)
Dataset of a subject and modeling data for an auditory word identification task.
data(word_classification_data)data(word_classification_data)
Data from the four experiments and model estimates
ExperimentNumberExperiment identifier
PresentationMethodMethod of presentation in the experiment: loudspeaker, headphones 3. Trial: Trial number in the experimental list
TrialScaledscaled Trial
Subjectanonymized subject identifier
Itemword identifier -german umlaute and special character coded as 'ae' 'oe' 'ue' and 'ss'
ActivationNDL activation
LogActivationlog(activation+epsilon)
L1normL1-norm (lexicality)
LogL1normlog of L1-norm
RecognitionDecisionrecognition decision (yes/no)
RecognitionRTlatency for recognition decision
LogRecognitionRTlog recognition RT
DictationAccuracydictation accuracy (TRUE: correct word reported, FALSE otherwise) 15. DictationRT: response latency to typing onset
Denis Arnold, Fabian Tomaschek, Konstantin Sering, Florence Lopez, and R. Harald Baayen (2017). Words from spontaneous conversational speech can be recognized with human-like accuracy by an error-driven learning algorithm that discriminates between meanings straight from smart acoustic features, bypassing the phoneme as recognition unit PLoS ONE 12(4):e0174623. https://doi.org/10.1371/journal.pone.0174623