Skip to contents

Track fundamental frequency and periodicity using CREPE (ONNX)

trk_pitch_crepe.Rd

Estimates F0 by applying a deep convolutional neural network directly to the time-domain waveform (CREPE; (Kim et al. 2018) ). Two model sizes are available: "tiny" (~1.9 MB, fast) and "full" (~85 MB, more accurate). No Python or PyTorch required — inference uses the bundled ONNX Runtime.

Usage

trk_pitch_crepe(listOfFiles, beginTime = 0, endTime = 0, windowShift = 10, windowSize = 15, minF = 50, maxF = 550, voicing.threshold = 0.21, silence.threshold = -60, model = c("tiny", "full"), decoder = c("viterbi", "argmax"), batch_size = 512L, explicitExt = "crp", outputDirectory = NULL, toFile = TRUE, verbose = TRUE)

Arguments

listOfFiles

Character vector of audio file paths. Any format supported by av is accepted.

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 10 ms (100 Hz).

windowSize

Numeric. Smoothing filter window size in milliseconds, applied to both median (periodicity) and mean (F0) post-processing filters. Default 15 ms.

minF

Numeric. Minimum F0 in Hz. Frames with estimated F0 below this value are set to 0 (unvoiced). Default 50 Hz.

maxF

Numeric. Maximum F0 in Hz. Frames above this value are set to 0. Default 550 Hz.

voicing.threshold

Numeric. Periodicity threshold for voicing decisions. Frames with periodicity below this value are set to unvoiced (F0 = 0). Default 0.21.

silence.threshold

Numeric. A-weighted dB threshold relative to global maximum. Frames below this are treated as silent (periodicity = 0). Default -60 dB.

model

Character. Model variant: "tiny" (fast) or "full" (more accurate). Default "tiny".

decoder

Character. Decoding method: "viterbi" (recommended, smoother) or "argmax" (faster but noisier). Default "viterbi".

batch_size

Integer. Number of frames per ONNX inference batch. Default 512.

explicitExt

Character. Output file extension. Default "crp".

outputDirectory

Character. Directory for output files. NULL (default) writes alongside the input file.

toFile

Logical. If TRUE, write SSFF output files and return the count written (invisibly). If FALSE, return an AsspDataObj. Default TRUE.

verbose

Logical. Print per-file progress. Default TRUE.

Value

If toFile = FALSE: an AsspDataObj with tracks:

f0

REAL32, Hz, n\_frames × 1. Fundamental frequency; 0 in unvoiced/silent frames.

periodicity

REAL32, 0–1, n\_frames × 1. Model confidence; values below voicing.threshold are treated as unvoiced.

Frame rate: 1000 / windowShift Hz (default 100 Hz, 10 ms hop). If toFile = TRUE: integer count of files written, returned invisibly.

Details

ONNX Runtime is installed automatically on first use (~30 MB, cached in the R user directory) and persists across R sessions and package reinstalls. The model file (tiny: ~2 MB, full: ~85 MB) is downloaded from the torchcrepe-onnx Hugging Face Hub repo on first use (requires the huggingfaceR package and a network connection) and cached in the R user directory; subsequent calls read the cached copy with no network access.

Post-processing (matching torchcrepe): median filter on periodicity → A-weighted silence detection → voicing threshold → NaN-aware mean filter on F0 → F0 range clamping.

References

Kim JW, Salamon J, Li P, Bello JP (2018). “Crepe: A Convolutional Representation for Pitch Estimation.” 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 00, 161–165. doi:10.1109/icassp.2018.8461329 .

Examples

if (FALSE) { # \dontrun{
trk_pitch_crepe(
  system.file("samples", "sustained", "a1.wav", package = "superassp"),
  toFile = FALSE
)
} # }