I’m trying to get a decibel reading from an audio stream using the sounddevice library.

import sounddevice as sd
import numpy as np

class SoundLevelReader(QObject):
    new_level = Signal(float)
    def sd_callback(self, indata, frames, time, status):
        # see https://en.wikipedia.org/wiki/Sound_pressure#Sound_pressure_level
        rms = np.sqrt(np.mean(np.absolute(indata)**2))
        ref = 20e-6 # reference sound pressure in air
        db  = 20 * np.log10(rms / ref)
        print("RMS:{:.6f}".format(rms))
        print("REF:{:.6f}".format(ref))
        print("RES:{:.6f}".format(db))
        self.new_level.emit(db)

def main():
    # ...  
    reader = SoundLevelReader()     
    with sd.InputStream(callback=reader.sd_callback, device=4, blocksize=5000):
        sys.exit(app.exec())

I’m basically expecting decibel values in a range from -60 to 0 (theoretically).

However, by trying to respect the reference value of 20µP (ref), the resulting values get way too high – even way into the positive:

• db = 20 * np.log10(rms / ref)

RMS:0.000818
REF:0.000020
RES:32.236420

RMS:0.000455
REF:0.000020
RES:27.131226

But if I leave out the reference entirely, values seem to be more trustworthy:

• db = 20 * np.log10(rms)

RMS:0.013473
REF:0.000020
RES:-37.410866

RMS:0.003492
REF:0.000020
RES:-49.137325

The funny thing is that I only get the same values like my DAW, when I use 10 instead of 20 as a factor:

• db = 10 * np.log10(rms)

RMS:0.002143
REF:0.000020
RES:-26.689749

RMS:0.033025
REF:0.000020
RES:-14.811516

Question:

Can someone explain, how to correctly calculate a decibel value based on readings coming from the sounddevice python lib? I know there are many different decibel “variants”. My main goal is to get a reading that’s exactly how DAWs these days would display it.