Impulse Response Measurer: Amplitude of recorded signal is almost half compared to recording in audacity

Hi,

i have recently been working with the impulse response of a room and wanted to determine it myself without the impulseresponsemeasurer, using the results of the tool as a guide for my own attempts.

I have now noticed that the recorded values from the irdata table are about half the size of the recorded values from audacity.

I use the saved signal to be sent from the irdata table and play it back in audacity, while simultaneously recording in audacity. Playback volume is the same in both cases, recording volume is also the same.

I can't explain now how these differences come about, since the same hardware is used.

In the picture you can see that the recorded signals are the same in their characteristics - only the values for the amplitude do not match. I am not sure now if I get wrong values for the impulse response in the impulseresponsemeasurer, because the ratio of the signal to be played and the recorded signal do not fit...

Is this how it should be?

Have I overlooked something in the documentation?

Am I on the wrong track? Are the values in audacity wrong?

Is there a hidden option/value for the recording volume in matlab which affect the recorded values?

If someone could make similar observations or give me a tip how these deviations come about I would be grateful.

Greetings,

Mehmet

Answer 1

[Jimmy Lapierre]

Playback level in the impulseResponseMeasurer is controlled by the "Excitation Level (dBFS)" setting in the toolbar. Does setting that to 0 dBFS get you more in line with what you expect?

[Mehmet Cibir]

Hi Jimmy,

the Exitation Level is already at 0dBFS so that the to played signal as an amplitude of 1,-1. Thats also the value which i can see in Audacity if i export the data from Matlab as .wav file. If i change it to -6dBFS ( 1/2 amplitude i guess ) the ecitated and recorded signal has the half amplitude compared to the 0dbFS measurement.

However, i get in the impulseresponsemeasurer as well as via a written script using the audioPlayerRecorder function only the (almost) half amplitude. While the Analog Input Recorder gives me the same values as Audacity when I play the sweeptone, which in turn looks the same in Matlab as in Audacity, so the same start conditions.

I briefly thought it might be the interpretation of mono/stereo signals, but i discarded that because i have to specify the player and recording channel in both cases.

The script i used instead of the impulseresponsemeasurer is attached below. I could simply double the recorded values, but i think thats only a workaround and solves not the rootcause....

% impulse of a room

%% Erstelle Sinesweep

FS = 48000; % your sampling rate

exc = sweeptone(5, 0.5, FS, 'SweepFrequencyRange', [20 20000],'ExcitationLevel',0); %%

excrep = repmat(exc,3,1); % duplicate the sweep to reduce the noise floor

audiowrite("sweep5-05-20-20000.wav",excrep,FS); % create file for playback on site

plot((1:length(excrep))/FS, excrep)

xlabel('Time (seconds)')

ylabel('Amplitude')

%% Record and play

fileReader= dsp.AudioFileReader('sweep5-05-20-20000.wav','SamplesPerFrame',1024);

fs = fileReader.SampleRate;

fileWriter = dsp.AudioFileWriter('recorded-sweep5-05-20-20000.wav','SampleRate',fs);

aPR = audioPlayerRecorder('SampleRate',fs,'PlayerChannelMapping',3,'RecorderChannelMapping',5,'BitDepth','16-bit integer');

while ~isDone(fileReader)

audioToPlay = fileReader();

[audioRecorded,nUnderruns,nOverruns] = aPR(audioToPlay);

fileWriter(audioRecorded)

if nUnderruns > 0

fprintf('Audio player queeu was underrun by %d samples.\n',nUnderruns);

end

if nOverruns > 0

fprintf('Audioplayer queue was overrun by %d samples.\n',nOverruns);

end

end

release(fileReader)

release(fileWriter)

release(aPR)

%%

[rec_sweep,fs] = audioread("recorded-sweep5-05-20-20000.wav"); % load the recording

%% for trimming of recorded sweep

posStart = length(rec_sweep)-length(exc)+1

posEnd = length(rec_sweep)

%%

rec_sweep_cuted = rec_sweep(posStart:posEnd,1);

%%

ir = impzest(exc,rec_sweep_cuted);

plot((1:length(ir))/fs, ir)

xlabel('Time (seconds)')

xlim([0 1]) % zoom on the beginning (increase this for actual recording)

title('Room impulse response (time)')

%%

IR = fft(ir);

semilogx(linspace(0,fs/2,length(IR)/2+1), db(abs(IR(1:end/2+1))))

xlim([20 20000]), grid on

title('Room impulse response (frequency)')

xlabel('Frequency (Hz)')

ylabel('Amplitude (dB)')

[Jimmy Lapierre]

Would you be able to compare the levels with audiorecorder and/or audioDeviceReader ?

Does changing the BitDepth option make a difference?

Just FYI, if you want, you could read the excitation from memory to reduce the odds of dropouts instead of writing the sweep file (same with recording).

excdata = dsp.AsyncBuffer(size(excrep,1));
write(excdata, excrep);
%...
audioToPlay = read(excdata,1024);

I also noticed one thing: you want to use your 3 reps:

REP = 3;
excrep = repmat(exc,REP,1);
%...
posStart = length(rec_sweep)-REP*length(exc)+1;

[Mehmet Cibir]

Hi Jimmy,

thank you for your input.

I made a quick and dirty measurement with audiorecord and audiodevicereader and there the amplitudes are exactly as big as in audacity, or twice as big as in audioplayerrecorder/impulseresponsemeasurer.

I also played around with different bit depths (16,24,32-bit) and could not find any differences.

I think i will rewrite my script a bit and use a combination of audiodeviceader/audiodevicewriter.

But i still don't know where the differences come from. is it a bug or am i just missing something?

Greetings,

Mehmet

[Jimmy Lapierre]

The audioPlayerRecorder uses audioDeviceWriter and audioDeviceReader under the hood, so those should match. Can you let me know what version of MATLAB you are using and maybe share your code creating these objects.

[Mehmet Cibir]

Ah ok - good to know :-)

I'm using R2021b - my code for the quick'n'dirty measurement (audio was played by Audacity):

% test with audiorecorder

recorder = audiorecorder(48000,16,1,1);

disp('Aufnahme begonnen')

recordblocking(recorder,20)

disp('Aufnahme beendet')

Record_from_audiorecorder = getaudiodata(recorder);

%%

% test with audioDeviceReader

deviceReader = audioDeviceReader(48000,1024,"BitDepth","16-bit integer",'Device','Externes Mikrofon (Realtek(R) Audio)');

setup(deviceReader);

fileWriter = dsp.AudioFileWriter("rec_sweep_from_audio_dev_reader.wav",FileFormat="WAV");

disp('Aufnahme beginnt')

tic

while toc < 10

acquiredAudio = deviceReader();

fileWriter(acquiredAudio);

end

disp('Recording complete.')

release(deviceReader)

release(fileWriter)

[Jimmy Lapierre]

So the difference is Audacity is playing. What happens if you use audioPlayerRecorder to record, but play to a dummy channel (no speaker,etc) and play on the correct channel with Audacity like you are doing for this code?

[Mehmet Cibir]

Hi,

i tried your suggestion and found out that it is not that easy because audiacity tells me that the audio host or the audio device is blocked when you want to play the audio but the script is already running that uses audioPlayerRecorder.

It works if first the audio file is played in audacity and then the audioPlayerRecorder function is used ( or the script is executed ), but for this i had to use my external usb soundcard to play the audio file ( recording is with the internal soundcard as before)

The amplitude of the recording from the audioPlayerRecorder seems a bit higher, but on the whole only half as much as i would expect.

Due the changed constellation this can not really be compared with each other because now I also use another sound card for playback and the amplifier may behave differently than that of the internal sound card.

I'll try again tomorrow to make a comparison in audacity and matlab only with my external soundcard - just to exclude that it is somehow related to the internal soundcard/driver and i'm chasing problems where there are none....

Greetings,

Mehmet

[Mehmet Cibir]

Hi,

after a night of restful sleep and a reboot of my laptop, i gave it another try.

What can i say, today the recorded amplitudes in MATLAB (script/impulsresponsemeasurer) and Audacity match.

I don't know what the reason could be - in the end it was probably a layer 8 problem - sorry if i wasted your time!

Thanks again for your support!

Greetings,

Mehmet

[Jimmy Lapierre]

No problem, glad it worked out!

Answer 2

[Image Analyst]

I had the opposite situation. In audacity, it recorded a signal and said it was going from -0.5 to +0.5 with lots of clipping at -0.5 and +0.5 (which may have been caused by the dictation recorder I had used to record the conversation). Then I told Audacity to export it to .mp3 format and read it into MATLAB and now the values go from -0.7 to +0.7. The waveform doesn't have flat tops like in Audacity so I think that those might have been changed upon conversion to MP3 format, which changes the signal due to lossy compression.

What format did you export the signal to from Audacity?

[Mehmet Cibir]

Hi,

i exported the ExitationSignal from the irdata_xxx table as .wav file with audiowrite(...), also the RecordedSignal to compare it with the recorded signal in Audacity.

Then i exported my recorded signal (track) as .wav (Signed 16-bit PCM) and used "[y,FS] = audioread("rec_signal_from_audacity")" to check if there is any changes in amplitude because of conversion or some black magic :-)

The image below clearly shows that there is no conversion of the amplitude by exporting. The recorded signal in the impulseresponsemeasurer tool has about half the amplitude as the recorded signal/track in Audacity.

The only other thing I could do would be to repeat the measurements in Audacity and Matlab and measure the voltages at the microphone with an oscilloscope. If the voltage levels are the same, I would conclude that there was a conversion/interpretation error when saving the values in one of the two programs.

Answer 3

[Mehmet Cibir]

I tried again and measured the microphone voltage while recording in Audacity and in MATLAB with the audioPlayerRecorder() function. The voltage levels are the same, which should lead to the same results in the recording volume.

The recorded amplitude is with the impulseresponsemeasurer and the audioPlayerRecord function the same, and both are about the half compared to the recorded from Audacity.

But if i try to record with the Analog Input Recorder, the amplitude is the same as in Audacity, see picture below.

In my opinion, what was recorded by Audacity and the Analog Input Recorder is correct in terms of amplitude.

But why MATLAB makes differences between the Analog Input Recorder and the audioPlayerRecorder/impulseresponsemeasurer is a mystery to me.