Skip to contents

Detect Silences in Sound and Create TextGrid

Source: R/vad.R
sound_to_textgrid_silences.Rd

Detects silent and sounding (voiced) intervals in a Sound object and returns a TextGrid with labeled intervals. This is essential for AVQI calculation, which requires extraction of voiced segments from continuous speech.

Usage

sound_to_textgrid_silences(
  sound,
  minimum_pitch = 100,
  time_step = 0,
  silence_threshold = -25,
  min_silent_interval = 0.1,
  min_sounding_interval = 0.1,
  silent_label = "silence",
  sounding_label = "sounding"
)

Arguments

sound

Sound object to analyze

minimum_pitch

Numeric. Minimum pitch for intensity calculation (Hz, default: 100)

time_step

Numeric. Time step for intensity calculation (s, default: 0.0 = auto)

silence_threshold

Numeric. Silence threshold in dB below maximum (default: -25)

min_silent_interval

Numeric. Minimum duration of silent interval (s, default: 0.1)

min_sounding_interval

Numeric. Minimum duration of sounding interval (s, default: 0.1)

silent_label

Character. Label for silent intervals (default: "silence")

sounding_label

Character. Label for sounding intervals (default: "sounding")

Value

TextGrid object with one interval tier containing silent and sounding intervals

Details

The function works by: 1. Computing the intensity contour of the sound 2. Identifying regions where intensity falls below `silence_threshold` dB relative to the maximum intensity 3. Merging nearby silent/sounding regions based on minimum duration criteria 4. Creating a TextGrid with labeled intervals

For AVQI, use these parameters (matching Praat AVQI script): - `minimum_pitch = 50` - `time_step = 0.003` - `silence_threshold = -25` - `min_silent_interval = 0.1` - `min_sounding_interval = 0.1`

Examples

# Synthetic speech-like sound: loud / near-silent / loud
sound <- sounds_append(
  sounds_append(
    Sound$create_tone(frequency = 200, duration = 0.5, amplitude = 0.8),
    Sound$create_tone(frequency = 200, duration = 0.3, amplitude = 0.001)
  ),
  Sound$create_tone(frequency = 200, duration = 0.5, amplitude = 0.8)
)

# Create TextGrid with voice activity detection
vad_grid <- sound_to_textgrid_silences(
  sound,
  minimum_pitch = 50,
  time_step = 0.003,
  silence_threshold = -25,
  min_silent_interval = 0.1,
  min_sounding_interval = 0.1
)

# Extract voiced intervals
voiced_intervals <- textgrid_get_intervals_where(
  vad_grid,
  tier = 1,
  condition = "equals",
  text = "sounding"
)

# Extract voiced parts
voiced_sounds <- sound_extract_parts(
  sound,
  voiced_intervals$xmin,
  voiced_intervals$xmax,
  window_shape = "rectangular",
  relative_width = 1.0,
  preserve_times = FALSE
)