Here, we are comparing the output sound quality of the following devices :
*
Nagra Mezzo,
*
Tascam DR05 V1.
For this, we simply made simultaneous field recordings of the same audio scenes
(soundscapes) by starting all the recorders at the same times.
With some various settings each time.
All recordings have been made :
* at 48kHz sampling rate,
* some as 24 bit-WAV, some as 320 kbps MP3.
Then, they were :
1. loaded into the
Audacity software,
2. cut down to a special time range,
* for selecting the target signal event (often a bird call or song note),
* for device to device time matching,
3. and finally converted to 32 bit-WAV files.
All in all, they suffered :
* no post-processing,
* exactly the same preparation steps.
For every audio scene, you'll find :
* the date and time of recordings,
* the place where it was recorded (in Auvergne region, center-south of France),
* some words about the audible sound sources (birds, human activities, wind, ...),
* for each recording :
- the name of the device used,
- the device settings used for the recording (see the abbreviations used below),
- a link to the original audio file analysed (for downloading if needed :
right-click, then 'save link target'),
- a player to listen to an MP3 (VBR 3) 0 dB RMS normalised version
of the analysed recording (to get a decent volume).
* 3 spectrograms for each recording :
- a raw one (only "amplified" through a multiplication by 50, for better constrast ;
unit = power spectral density),
- a "normalised" one (reference = max. of the spectrograms of all the recordings ;
unit = dB, in some kind of Full Scale variant, but the "full scale" is here that
of the actual sample values, not the one of the max possible value for 24bit samples),
- a "SNR" = Signal to Noise Ratio one, computed by dividing the whole raw one
by the temporal mean of the raw one during the "noise" time range (see below) ;
unit = dB.
- note: the orange box on each spectrogram shows the time range selected
for extracting the "pure" noise average (approximately no audible signal inside).
* 3 PSDs plots (power spectral density, extracted from the raw spectrograms) for each recording :
- a "normalised" one (reference = max. of the raw spectrograms of all the recordings,
but only during the "signal" time range, and only in the "signal" frequency band,
highlighted through 2 pink boxes on the spectrograms ; unit = dB, like the norm'ed spectrogram),
- the device by device difference (of the normalised PSD) to the 1st one,
- a "SNR" = Signal to Noise Ratio one, computed by dividing the temporal mean
of the raw spectrogram during the "signal" time range (pink box),
by the temporal mean of the same spectrogram during the "noise" time range (orange box) ;
unit = dB,
* some comments after analysing the spectrograms and PSDs, as well as listening to the recordings
(note: for "ear" = subjective comparison, some simple amplification was achieved,
in order to get roughly the same "perceived" signal volume, = bird song or call volume).
Note: The simple signal processing computations (spectrogram, PSD, amplification, division, mean, ...)
are run through
Python 3.5 code,
using
SciPy 1.1
and
NumPy 1.14 libraries
(see the last line of this page for more details and access to the source code).
Device settings abbreviations (note: on/off settings are all 'off' unless specified) :
- r : sample rate (kHz) ex: r=48
- wav : recorded in Wave format, with this bit depth / resolution (bit). Ex : wav=24
- mp3 : recorded in MP3 format, with this bit rate (bps). Ex : mp3=320
- g : mic. gain (% of the possible range, or dB, or a=auto=agc). Ex: g=50%, g=16dB, g=a
- hn : device in hand (on/off). Ex hn=on
- mp : electronic microphone pad (on/off). Ex mp=on
- ws : physical windscreen (on/off). Ex: ws=on
- mpw : microphone power (on/off). Ex: mpw=on
- hpf : high-pass filter, with given threshold (Hz). Ex: hpf=200
Finally, for this device comparison,
* only device-embedded microphones where used (no external one),
* care was taken to avoid any dust or other obstacle around the microphones,
* the 'external microphone power' setting was always set to off.
And last precision : the latest available firmware (1.3.1.8) had been installed
on the Nagra Mezzo before starting the field recordings.
The above described parallel field tests and objective analyses
show some contrasting results when comparing the Nagra Mezzo to the Tascam DR05 v1.
But all in all, the Tascam DR05 v1 very often "displays" :
* a better signal to noise ratio (+1 to +2dB),
* far less low-frequency noise below 300Hz (-1 to -14dB),
* less medium-frequency noise below 2kHz (-1 to -4dB),
* a slighltly better medium to high frequency lines separation,
* a better "ear feeling", mainly due to less LF noise.
The Mezzo windscreen is - not surprisingly - quite efficient at lowering the low frequency noise
... when the wheather is windy. But the effect would be same on the DR05.
During these tests, the high pass filter option (no more than the auto. gain control one)
was not activated ... as I don't want to hide any possible low frequency signal at recording time
(of course, I often apply this kind of filter ... when post-processing the recordings,
when I'm sure it won't destroy any interesting detail).
For some background about me,
* although I'm not an audio signal processing expert, I'm a computer science engineer
and thus know more than some basics about it, as well as about sound physics,
* as an active nature watcher, more specificaly dedicated to ornithology,
I also have a long experience and passionnate practice of nature soundscapes live analysis by ear,
* and thus an excellent knowledge of the various calls and songs of most of the species
of birds that can be encountered at least in the center of France ;
* thanks to this, among others,
- I have been an active contributor to the french
Museum National d'Histoire Naturelle common birds monitoring program called
STOC EPS since 2002,
- I'm also an active contributor of the
LPO Auvergne Distance Sampling
group created in 2016 by
François Guélin,
- I've been providing since 2015 serious training courses, applied field sessions and online exercises
about recognising common bird species by ear (see my
training, reference and exercice material (in french) about this),
* I've come to recording nature soundscapes only some years ago,
as a non-top-priority activity, targetting more convenience and pocket handling than high fidelity,
* but I felt recently a need to improve the sound quality of my recordings.