Track formant frequencies and bandwidths (FOREST)

Estimates vocal tract resonance (formant) frequencies and their bandwidths using the FOREST algorithm from the libassp C library (Scheffers 2012) . Root-solving of the LP polynomial is guided by Pisarenko frequencies from the Split-Levinson Algorithm (SLA) to classify resonances as formants. Prefer trk_formant_forest when broad compatibility and speed on large corpora are priorities.

Usage

trk_formant_forest(listOfFiles, beginTime = 0, endTime = 0, windowShift = 5, windowSize = 20, effectiveLength = TRUE, nominalF1 = 500, gender = "m", estimate = FALSE, order = 0, incrOrder = 0, numFormants = 4, window = "BLACKMAN", preemphasis = -0.8, toFile = TRUE, explicitExt = "fms", outputDirectory = NULL, assertLossless = NULL, logToFile = FALSE, convertOverwrites = FALSE, keepConverted = FALSE, verbose = TRUE)

Arguments

listOfFiles: Character vector of audio file paths. Any format supported by av is accepted; non-native inputs are transcoded automatically.
beginTime: Numeric. Start of analysis window in seconds. Default 0 (file start).
endTime: Numeric. End of analysis window in seconds. Default 0 (file end).
windowShift: Numeric. Frame shift in milliseconds; sets output frame rate (1000 / windowShift Hz). Default 5 ms.
windowSize: Numeric. Analysis window size in milliseconds. Default 20 ms.
effectiveLength: Logical. Make window size effective rather than exact. Default TRUE.
nominalF1: Numeric. Assumed F1 frequency in Hz used to build the formant range table. Increase by ~12% for female voices. Default 500.0 Hz.
gender: Character. Gender-specific parameter preset: "f" (female, sets window to 12.5 ms and nominalF1 to 560 Hz), "m" (male), or "u" (unknown, default).
estimate: Logical. Insert rough frequency estimates for missing formants rather than returning zero. Default FALSE.
order: Integer. Decrease the default LPC filter order by 2 (one fewer resonance). Default 0 (no change).
incrOrder: Integer. Increase the default LPC filter order by 2 (one more resonance). Default 0 (no change).
numFormants: Integer. Number of formants to track (maximum 8 or half the LPC order). Default 4.
window: Character. Analysis window function type. Default "BLACKMAN". See AsspWindowTypes.
preemphasis: Numeric. Pre-emphasis factor (-1 <= val <= 0); default is sample-rate- and nominalF1-dependent.
toFile: Logical. If TRUE, write SSFF output files and return the count written (invisibly). If FALSE, return an AsspDataObj. Default TRUE.
explicitExt: Character. Output file extension. Default "fms".
outputDirectory: Character. Directory for output files. NULL (default) writes alongside the input file.
assertLossless: Character vector of additional file extensions to treat as losslessly encoded.
logToFile: Logical. Write processing log to a file in outputDirectory rather than the console. Default FALSE.
convertOverwrites: Logical. Allow transcoding to overwrite existing files. Default FALSE.
keepConverted: Logical. Retain intermediate transcoded files. Default FALSE.
verbose: Logical. Print per-file progress. Default TRUE.

Value

If toFile = FALSE: an AsspDataObj with tracks:

F[Hz]: REAL32, Hz, n_frames x numFormants columns. Estimated centre frequency of each formant (0 = missing).
B[Hz]: REAL32, Hz, n_frames x numFormants columns. Bandwidth of each formant.

Frame rate: 1000 / windowShift Hz (default 200 Hz). If toFile = TRUE: integer count of files written, returned invisibly.

Details

The gender preset overrides windowSize and nominalF1. Set estimate = TRUE to fill missing formant slots with rough estimates rather than zeros, which can help downstream processing.

Author

Raphael Winkelmann

Lasse Bombien

Fredrik Nylén

Examples

# get path to audio file
path2wav <- list.files(system.file("samples","sustained", package = "superassp"), pattern = glob2rx("a1.wav"), full.names = TRUE)


# calculate formant values
res <- trk_formant_forest(path2wav, toFile=FALSE)
#> Applying `method(trk_formant_forest, class_character)()` to 1 recording

# plot formant values
matplot(seq(0, n_records(res) - 1) / sample_rate(res) +
          attr(res, 'startTime'),
        res[["F[Hz]"]],
        type='l',
        xlab='time (s)',
        ylab='Formant frequency (Hz)')