I wanted to record some voice notes, so I bought an inexpensive USB computer headset. It was OK, but between words, where there should have been silence, there was a lot of white noise; hiss, as produced by old audio technology from the last century. So I bought another headset, a famous name brand, at twice the price. It had more noise! Thus began a quest to find a decent, inexpensive USB headset.
I developed a test plan, and then borrowed headsets from friends. This page describes what I found.
Each microphone is positioned about 3 cm from the front of a sound system speaker, facing the speaker. In Windows 7, the mic volume is set at 100. Using Audacity 1.3.13, with mic volume at 1.0, the volume of the stereo system speaker is adjusted until the Audacity microphone level indicator registers –12 dB during white noise (from http://en.wikipedia.org/wiki/White_noise). Several seconds of white noise are recorded, then several seconds of silence (with the stereo system powered off). The room is very quiet to the ear, but there is probably lower level noise present.
Fig 1. A typical recording of white noise followed by quiet room 'silence'
The white noise and silent segments are analysed with Sonic Visualizer 1.9 from the University of London. A 'spectrum layer' is added, with the default settings and a window size of 65536, to display the signal power versus frequency.
The white noise input file (referenced above) has the following spectrum as displayed by Sonic Visualizer:
Fig 2. Power spectrum of white noise, with 'piano keyboard' labeling frequencies
The x-axis is frequency, from 0 to 48 KHz. The scale is logarithmic. The 'piano keyboard' at the bottom shows where the frequencies of piano keys are on the scale, with the tiny blue square marking the middle C key (440 Hz). This keyboard has more octaves than most keyboards; the left end key is about 8 Hz, and the right end key is about 11.5 KHz.
The y-axis is power. An ideal white noise signal would have equal power across the spectrum.
Below are the results for three microphones (two high-quality digital recorders, the Zoom H2n and H4n, and an inexpensive noise-cancelling microphone, the Olympus ME52-W) and several headsets: Plantronics .Audio 476 DSP, Logitech ClearChat Comfort USB, and the Cyber Acoustics AC-840.
The blue signal is an analysis of the white noise signal, and the orange is silence (thus shows how much self-noise is produced by the microphone itself, along with any quiet-room noise).
The Zoom H4n serves as a reference. According to its specifications, its internal microphones are quite flat over the range, and thus it is probably the best record of what the stereo speakers actually emitted when the white noise signal was output; similarly, it is probably a good indicator of what noise is present in the quiet room (though of course the H4n also has some self-noise).
The Zoom H4n shows good separation across the spectrum between white noise and the silence, and fairly low self-noise (plus room noise), at -50 dB. The Olympus ME52 has comparably low self-noise. However, the headsets generally don't perform as well. The Plantronics 476 has a noticeable hiss (noise to -40 dB). The Logitech ClearChat is quieter, but has a pronounced 60 Hz hum (the spike). The Cyber Acoustics AC840 attenuates the lower end of the spectrum, resulting in a 'nasal' effect for low-pitch male voices.
The Zoom H2n chart was generated a year later. I'm not sure if the excellent separation between white noise and silence can be fairly compared with the other charts; procedures and conditions may be different.
All the microphones recorded a 60 Hz signal to some extent, the Logitech ClearChat being most sensitive (resulting in an objectionable hum).
Subjectively, the voice quality of the Plantronics 476 headset was good with the least distortion, but there was significant 'hiss' or 'amplifier noise' in the silent parts between words. The Cyber Acoustics AC-840 seems quieter than the results indicate, but it imparts a 'nasal' effect for low-pitched voices. The voice quality of the Logitech ClearChat was good but the 60 Hz hum was impossible to overlook.
In earlier tests, the Plantronics .Audio 310 USB performed well, but it's now discontinued.
I would like to try Sennheiser headsets, but haven't yet had an opportunity.
I'm left with the impression that to obtain good voice quality without 'hiss' in the silent parts, I'll need to give up on inexpensive headsets.
For applications such as Skype, where there is already considerable background noise and/or limited transmission of lower frequencies, the Cyber Acoustics AC-840 or the Plantronics 476 would be fine; the Plantronics 476 has better voice quality but costs twice as much. The Logitech ClearChat's 60 Hz hum rules it out for me (also it's more expensive than the Cyber Acoustics AC-840). People with higher-pitched voices might find the AC-840 and 476 comparable.
I'd rank them as follows:
Cyber Acoustics AC-840: Best value, decent quality for VOIP, especially for people with high-pitched voices.
Plantronics .Audio 476 DSP: Decent quality for VOIP.
: 60 Hz hum picked up by the one I tested would be hard to overlook.
Sadly, I think none of them are good for quality voice recording, because of self-noise and/or distortion.
It'd be nice if the headset manufacturers would use better microphones. The Olympus ME-52W is only about $10, and it performed well.