Detecting silence from float audio values - audio

Question

I'm reading the audio data of a file as float, and I get for example these value:
-4,151046E+34
-2,365558E+38
6,068741E+26
-4,141856E+34
-2,179363E+38
1,177772E-04
-1,035052E+34
-1,#QNAN
2,668123E-20
-1,0609E+37
-2,153349E+38
1,105884E-16
-4,25223E+37
-1,#QNAN
-3,718855E+22
-1,695596E+38
I would like to detect when silence starts and ends.
Do these values represent values that are directly related to volume or would a value of 0 represent the value at this point in this screenshot, and I need to look at rather many of these values to detect silence?

Answer 1

Silence is a notion tied to perception which has an attribute of time ... silence cannot happen for just an instant of time surrounded by loud audio since it will not be perceived as silence
Silence happens when the audio curve is at or does not vary much from the zero crossing point for some perceivable period of time ... you cannot have listenable audio followed by silence which lasts for just an instant of time followed by listenable audio ... that is not silence ... your eardrum or the membrane of a microphone in a silent room does not vibrate ... as the loudness of a room increases from silence those surfaces begin to wobble ... the plot you show can be thought of as visualizing this wobble ... on the plot the only silence happens during that flat line period of time at the beginning
To programmatically identify when silence occurs you need two parameters
some maximum height of audio curve below which you declare silence happens
some minimum length of time the audio curve stays below that max height
You can experiment with guessing these values ... now lets identify when silence happens
package main
import "fmt"
func main() {
// somehow your audio_buffer gets populated
flag_in_candidate_silence := false // current sample is quiet
flag_currently_in_declared_silence := false // current stretch of samples are in silence period
total_num_samples := len(audio_buffer) // identify how many samples
max_vol := 0.1 // max volume and still a silence candidate
min_num_samples := 2000 // minimum number of samples necessary to declare silence has happened
// value used is dependent on sampling rate
curr_num_samples_found := 0
index_silence_starts := 0
index_silence_ends := 0
for curr_sample := 0; curr_sample < total_num_samples; curr_sample++ {
curr_amplitude := audio_buffer[curr_sample]
if curr_amplitude < max_vol { // current sample is candidate for silence
index_silence_ends = curr_sample
if flag_in_candidate_silence != true { // previous sample was not a candidate
index_silence_starts = curr_sample
}
if curr_num_samples_found > min_num_samples {
// we are inside a period of silence !!!!!!!!!!!
flag_currently_in_declared_silence = true
}
flag_in_candidate_silence = true
curr_num_samples_found++ // increment counter of current stretch of silence candidates
} else {
if flag_currently_in_declared_silence == true {
fmt.Println("found silence stretch of samples from ", index_silence_starts, " to ", index_silence_ends)
}
flag_in_candidate_silence = false
flag_currently_in_declared_silence = false
curr_num_samples_found = 0
}
}
if flag_currently_in_declared_silence == true {
fmt.Println("found silence stretch of samples from ", index_silence_starts, " to ", index_silence_ends)
}
}
(code not tested - spouted directly from forehead)