Skip to contents

Extract Voxit prosodic complexity features from audio files

lst_voxit.Rd

Computes voice and articulation complexity measures from audio files using word-level alignments (optional) and WORLD-based pitch/spectral analysis. Features include speaking rate (WPM), pause statistics, F0 dynamics, pitch complexity, and intensity.

Uses superassp's C++ implementations of WORLD vocoder analysis (Harvest F0, intensity, LZ complexity, SG smoothing) — no external dependencies beyond those already in superassp.

Usage

lst_voxit(listOfFiles, alignmentFiles = NULL, beginTime = 0, endTime = 0, minF = 60, maxF = 600, verbose = TRUE, parallel = TRUE, n_cores = NULL, toFile = FALSE, return_jstf = FALSE, explicitExt = "vxt", outputDirectory = NULL)

Arguments

listOfFiles

Character vector of audio file paths

alignmentFiles

Character vector of alignment CSV paths (optional). CSV must have columns: word, start, end (in seconds). If NULL, word-based metrics (WPM, pause stats, rhythmic complexity) are set to NA; signal-based metrics are always computed.

beginTime

Numeric. Start of analysis window in seconds (default: 0)

endTime

Numeric. End of analysis window in seconds (default: 0)

minF

Numeric. Minimum F0 in Hz for Harvest (default: 60)

maxF

Numeric. Maximum F0 in Hz for Harvest (default: 600)

verbose

Logical. Show progress (default: TRUE)

parallel

Logical. Use parallel processing (default: TRUE)

n_cores

Integer. Number of cores (default: auto-detect)

toFile

Logical. Write JSTF files (default: FALSE)

explicitExt

Character. Output extension (default: "vxt")

outputDirectory

Character. Output directory (default: NULL = input dir)

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

If toFile=FALSE (default): for single file, a named list with up to 18 prosodic features. For multiple files, a list where each element is a file's feature list. If toFile=TRUE, invisibly returns the output file path(s).

Features

Word-level (requires alignment):

  • WPM: Words per minute (speaking rate)

  • pause_count: Number of pauses (100–3000 ms)

  • long_pause_count: Pauses > 3 s

  • average_pause_length: Mean pause duration (s)

  • average_pause_rate: Pauses per second

  • rhythmic_complexity_of_pauses: LZ complexity of pause/speech sequence

Signal-level (WORLD-based):

  • average_pitch: Mean F0 (Hz) — arithmetic mean of geometric mean per voiced segment

  • pitch_range: F0 range in octaves

  • pitch_speed: F0 velocity (octaves/s)

  • pitch_acceleration: F0 acceleration (octaves/s²)

  • pitch_entropy: Shannon entropy of F0 histogram (bits)

  • f0_geometric_mean_hz: Geometric mean F0 (Hz)

  • voicing_percent: Percentage of voiced frames

  • intensity_mean_db: Mean intensity (dB)

  • lz_complexity_voiced: LZ complexity of VUV sequence

  • f0_velocity_mean_abs: Mean absolute F0 velocity (octaves/s)

  • f0_accel_mean_abs: Mean absolute F0 acceleration (octaves/s²)

  • dynamism: Composite measure: |f0MeanAbsVel| * f0Entropy + LZ * 0.439

Examples

if (FALSE) { # \dontrun{
# With alignment (all metrics)
features <- lst_voxit("speech.wav", alignmentFiles = "speech_align.csv")
print(features$WPM)
print(features$f0_geometric_mean_hz)

# Without alignment (signal metrics only)
features <- lst_voxit("speech.wav")
print(features$f0_geometric_mean_hz)  # OK
print(features$WPM)  # NA
} # }