Audio Recording - Testing Tesults

John Beale, 2004-2011

Here are a few measurements of the audio performance of digital recording equipment (and for fun, some codecs as well). The objective is to determine how clean a recording is possible from live analog sources, using various pieces of gear. The results might not be what you would expect. For example, the Sony VX2k records uncompressed, 48 kHz audio and it is about 10x as expensive as the Sharp MD recorder, but the Sharp MD has much better sound.

I have experimented with record levels to see if anything is significantly improved by different settings. I did find that it was a mistake to use a balanced->unbalanced audio transformer in the system (a box that converts a balanced signals, eg. from an XLR cable, to an unbalanced miniplug). The audio transformer caused severe distortion at +4 dB line level, and it was still very noticible at -10 dB. All of the results shown here are using direct connections with no transformers in the signal path.

To reduce the possibility of ground loops introducing 60 Hz hum, all recording devices were powered from batteries during the recording phase (except for the DPS-16). All playback was digital into the computer, either through USB (Neuros, iRiver, HHB minidisc), firewire (camcorders) or S/PDIF (DPS-16), so this is a test of the analog record mode only. On the Neuros and Zipi Z2 (only) I tested analog playback through the headphone output. The analog source for record measurements is the balanced line-output of the Echo Audio Mia soundcard, which is one of the cleaner soundcards available. 

Recording  hardware

Sony VX2000 and Sony TRV720 MiniDV and Digital8 camcorders
Panasonic DVX100 DVX100 audio   40Hz-15kHz 20-sec. Sweep,+4dBu (yes, it's real) Sweep,-10dBV
Panasonic TM700 AVCHD camcorder  higher input level  lower input level
Neuros (MP3 player) WAV recording -16dBV in, -10dBV in
Sony MZ-RH1 (line in) (mic in, 1k Z) (mic in, 150 ohm Z) (motor noise) HiMD minidisc recorder
HHB MDP500 and Sharp MD-DR7-A MD Recorders
Akai DPS-16 (HDD recorder) DPS16
24 bit, 96 kHz pro soundcard Echo Audio Mia
analog videotape (FM "HiFi" sound) VHS Tape
Sony HDR-FX1 HDR-FX1 camcorder in DV recording mode (line in)
CD-R to HHB DV343toHHB CD-R in DVD player through mixer to HHB (direct USB audio to PC)  
Sennheiser wireless EW 100 G2 wireless UHF wireless mic Tx/Rx set
iRiver iFP890 (MP3 recorder)  line level,  mic level
iRiver iFP895 (MP3 recorder) line level,  mic level
iRiver FP895 vs FP890 noise floor mp3 (resistor @ input, mic level 45, +40 dB in Cool Edit)
iRiver FP890 mic input noise floor mp3 (resistor @ input, mic level 55, +36 dB in CoolEdit)

Playback hardware

Neuros (MP3 player) output WAV Playback
Zipit Z2 headphone output  Z2_headphone  (mplayer on 44.1kHz 16bit wav file)


A Good Transformer

Doing event video, I often find there is a ground problem somewhere that puts hum on the audio feed from a house soundboard that I connect to the camera. You can generally fix this with an isolation transformer, but I wanted to ensure that the transformer didn't add too much distortion. As you can see from the measurement below comparing a signal chain both with and without the Jensen PI-2xx audio transformer, this particular model adds essentially no measurable distortion to the signal (at least that you can see with a 16-bit A/D converter).

Isolation transformer Jensen PI-2xx (a comparison with & without transformer in signal path)

For completeness, I also include additional plots for both left and right channels with PI-2xx and without PI-2xx. I used different types of cable for the left and right channel between the PI-2xx out and the JB930 line in, which may account for the frequency response. Here is a purely digital plot for DV343 SPDIF optical out.

Software Codecs

WAV  ( original 16/44.1 WAV test signal )
MP3 128k ( Cool Edit 2000 1.1 build 2418 )
Ogg Vorbis Q6  ( OggEnc v1.0.1, libvorbis 1.0.1, encoding -q6 )
AC3 192k  ( Vegas 4.0 192 kbps stereo, AC3ACM v0.7 decode )
AAC 128k  ( iTunes v4.5.0.31 AAC 128 kbps )
WMA 128k  (Vegas 4.0 128 kbps stereo, Window Media Audio ver.9 )
YouTube_LiveVideo  (YT: MPEG-2 layer 3 at 22.05k 46 kb/s mono, LV: MP3 at 44.1k 96 kb/s stereo)

There is no hardware test in this section, I simply encoded the test signal WAV file to each format, then decoded back to WAV and analysed the result. I also ran the analyser on the original WAV test file, to demonstrate the theoretical limit of a 16-bit file.  This test measures numbers that are useful to characterize analog systems, but  it is an open question how useful it is for lossy perceptual-based codecs. For example there is wide agreement that Ogg -Q6 sounds better on real music files than MP3 at 128k, but the MP3 file has better numbers on this test. The largest difference between codecs seems to be the high-frequency cutoff and the intermodulation distortion (IMD). This IMD test measures how well two different pure tones (here, 60 Hz and 7 kHz) can be reproduced, without generating distortion or noise at other frequencies.

The AC3 encoder or decoder that I used seems to have AGC action built in. Notice on the graph how the "dynamic range" pilot tone appears at -50 dB, but it is supposed to be at -60 dB, and it is with all the other codecs.

Back to main page.