Skip to contents

SEDREAMS Glottal Closure Instant Detection

lst_covarep_gci_sedreams.Rd

Detects glottal closure instants (GCIs) using SEDREAMS algorithm. Returns event times (GCI instants), not a regular frame grid.

Usage

lst_covarep_gci_sedreams(listOfFiles, beginTime = 0, endTime = 0, f0mean = NULL, polarity = NULL, return_jstf = FALSE, verbose = TRUE)

Arguments

listOfFiles

Vector of file paths (WAV, MP3, MP4, etc.) to analyze

beginTime

Start time in seconds (0 for beginning of file)

endTime

End time in seconds (0 for end of file)

f0mean

Estimated mean F0 in Hz. If NULL, auto-estimated from signal.

polarity

Signal polarity (1 or -1). If NULL, auto-detected.

verbose

Show progress messages (default: TRUE)

return_jstf

Logical. Return JsonTrackObj instead of data.frame? Default FALSE. When both toFile and return_jstf are TRUE, the file is written AND the object returned.

Value

Data frame with columns:

  • file: Input file path

  • n_gcis: Number of detected GCIs

  • gci_times: List column of numeric vectors (GCI times in seconds)

Details

SEDREAMS Algorithm (Ney and Kneser 2002) :

  1. Compute LPC residual (25ms frames, 5ms shift, order ≈ fs/1000 + 2)

  2. Bandpass filter signal around estimated F0 (mean-based signal)

  3. Find maxima/minima pairs in mean-based signal

  4. Locate GCI positions in LP residual peaks within windows

Typical output:

  • Voiced speech: 100-200 GCIs per second (F0-dependent)

  • Unvoiced/silence: 0 GCIs (no glottal closures)

Use cases:

  • Foundation for GCI-based voice quality (NAQ, QOQ, H1H2 via trk_covarep_vq_gci)

  • Voice pathology assessment (irregular GCI spacing = vocal pathology)

  • Glottal source analysis (GCI-anchored inverse filtering)

  • Speech analysis (pitch period estimation, voicing detection)

Downstream workflow:

  1. lst_covarep_gci_sedreams() — detect GCIs

  2. trk_covarep_vq_gci() — compute voice quality per GCI

  3. lst_covarep_vq() — summarize to scalars

References

Ney H, Kneser R (2002). “Speech recognition using continuous-space embeddings.” IEEE Signal Processing Magazine, 19(1), 33–42. Signal processing foundations for GCI detection and SEDREAMS algorithm.

Examples

if (FALSE) { # \dontrun{
# Single file
gcis <- lst_covarep_gci_sedreams("speech.wav", f0mean = 100)

# Batch process
files <- c("file1.wav", "file2.wav")
results <- lst_covarep_gci_sedreams(files, f0mean = 110)

# View results
results$gci_times[[1]]  # GCI times for first file (in seconds)

# Chain with voice quality analysis
vq <- trk_covarep_vq_gci("speech.wav", gci_times = gcis$gci_times[[1]])
} # }