Package 'audio.vadwebrtc'

Get from a Voice Activity Detection (VAD object) the segments which are voiced

Description

Postprocessing the Voice Activity Detection whereby sequences of voiced/non-voiced segments are collapsed by

1. first considering all non-voiced segments which are small in duration (default < 1 second) voiced

2. next considering voiced segments with length less than a number of seconds (default < 1 second) non-voiced

Usage

is.voiced(x, channel = 0, units = "seconds", ...)

Arguments

x	an object of class VAD as returned by VAD or VAD_channel
channel	integer with the channel, showing the voiced section of that channel only. Only used for segments extracted with VAD_channel
units	character string with the units to use for the output and thresholds used in the function - either 'seconds' or 'milliseconds'
...	further arguments passed on to the function

Value

A data.frame with columns vad_segment, start, end, duration, has_voice indicating where in the audio voice is detected

Examples

file   <- system.file(package = "audio.vadwebrtc", "extdata", "test_wav.wav")
vad    <- VAD(file, mode = "normal", milliseconds = 30)
vad$vad_segments
voiced <- is.voiced(vad, silence_min = 0.2, voiced_min = 1)
voiced
voiced <- is.voiced(vad, silence_min = 200, units = "milliseconds")
voiced

---

Voice Activity Detection

Description

Detect the location of active voice in audio. The Voice Activity Detection is implemented using a Gaussian Mixture Model from the "webrtc" framework. It works with .wav audio files with a sample rate of 8, 16 or 32 Khz an can be applied over a window of eiher 10, 20 or 30 milliseconds.

Usage

VAD(
  file,
  mode = c("normal", "lowbitrate", "aggressive", "veryaggressive"),
  milliseconds = 10L,
  type = "webrtc"
)

Arguments

file	the path to an audio file which should be a file in 16 bit with mono PCM samples (pcm_s16le codec) with a sampling rate of either 8Khz, 16KHz or 32Khz
mode	character string with the type of voice detection, either 'normal', 'lowbitrate', 'aggressive' or 'veryaggressive' where 'veryaggressive' means more silences are detected
milliseconds	integer with the number of milliseconds indicating to compute by this number of milliseconds the VAD signal. Can only be 10, 20 or 30. Defaults to 10.
type	character string with the type of VAD model. Only 'webrtc' currently.

Value

an object of class VAD which is a list with elements

• file: the path to the file

• sample_rate: the sample rate of the audio file in Hz

• channels: the number of channels in the audio - as the algorithm requires the audio to be mono this should only be 1

• samples: the number of samples in the data

• bitsPerSample: the number of bits per sample

• bytesPerSample: the number of bytes per sample

• type: the type of VAD model - currently only 'webrtc-gmm'

• mode: the provided VAD mode

• milliseconds: the provided milliseconds - either by 10, 20 or 30 ms frames

• frame_length: the frame length corresponding to the provided milliseconds

• vad: a data.frame with columns millisecond, has_voice and vad_segment indicating if the audio contains an active voice signal at that millisecond

• vad_segments: a data.frame with columns vad_segment, start, end and has_voice where the start/end values are in seconds

• vad_stats: a list with elements n_segments, n_segments_has_voice, n_segments_has_no_voice, seconds_has_voice, seconds_has_no_voice, pct_has_voice indicating the number of segments with voice and the duration of the voice/non-voice in the audio

Examples

file <- system.file(package = "audio.vadwebrtc", "extdata", "test_wav.wav")
vad  <- VAD(file, mode = "normal", milliseconds = 30)
vad
vad  <- VAD(file, mode = "lowbitrate", milliseconds = 20)
vad
vad  <- VAD(file, mode = "aggressive", milliseconds = 20)
vad
vad  <- VAD(file, mode = "veryaggressive", milliseconds = 20)
vad
vad  <- VAD(file, mode = "normal", milliseconds = 10)
vad
vad$vad_segments

## Not run: 
library(av)
x <- read_audio_bin(file)
plot(seq_along(x) / 16000, x, type = "l")
abline(v = vad$vad_segments$start, col = "red", lwd = 2)
abline(v = vad$vad_segments$end, col = "blue", lwd = 2)

##
## If you have audio which is not in mono or another sample rate
## consider using R package av to convert to the desired format
av_media_info(file)
av_audio_convert(file, output = "audio_pcm_16khz.wav", 
                 format = "wav", channels = 1, sample_rate = 16000)
vad <- VAD("audio_pcm_16khz.wav", mode = "normal")

## End(Not run)

file <- system.file(package = "audio.vadwebrtc", "extdata", "leak-test.wav")
vad  <- VAD(file, mode = "normal")
vad
vad$vad_segments
vad$vad_stats

---

Voice Activity Detection per channel

Description

Voice Activity Detection per channel. Transforms the audio file to a wav file with the provided sample_rate and perform the voice activity detection per channel.

Usage

VAD_channel(file, sample_rate = 16000, channels = c("default", "all"), ...)

Arguments

file	the path to an audio file
sample_rate	integer with the sample_rate to convert the file to. Passed on to av_audio_convert
channels	character string - either 'default' or 'all' indicating to do the voice activity detection for each channel independently ('default') or for all channels independently as well as all channels together ('all')
...	further arguments passed on to VAD

Value

an object of class webrtc-gmm-bychannel which is a list with elements

• file: the path to the file

• duration_secs: seconds

• sample_rate: the sample rate of the audio file in Hz

• channels: the number of channels in the audio

• samples: the number of samples in the data

• bitsPerSample: the number of bits per sample

• bytesPerSample: the number of bytes per sample

• type: the type of VAD model - currently only 'webrtc-gmm'

• mode: the provided VAD mode

• milliseconds: the provided milliseconds - either by 10, 20 or 30 ms frames

• frame_length: the frame length corresponding to the provided milliseconds

• vad_segments: a data.frame with columns channel, vad_segment, start, end and has_voice where the start/end values are in seconds

• vad_stats: a list with elements channel, n_segments, n_segments_has_voice, n_segments_has_no_voice, seconds_has_voice, seconds_has_no_voice, pct_has_voice indicating the number of segments with voice and the duration of the voice/non-voice in the audio

Channel 0 means all audio combined in 1 channel.

Examples

library(audio)
library(av)
file <- system.file(package = "audio.vadwebrtc", "extdata", "stereo.mp3")
vad  <- VAD_channel(file, sample_rate = 32000, 
                    mode = "normal", milliseconds = 10, channels = "all")
vad
vad$vad_segments
voiced <- is.voiced(vad, channel = 0, silence_min = 0.2, voiced_min = 1)
voiced
voiced <- is.voiced(vad, channel = 1, silence_min = 0.2, voiced_min = 1)
voiced
voiced <- is.voiced(vad, channel = 2, silence_min = 0.2, voiced_min = 1)
voiced

---