A Microphone Deep Dive: Part 1 – Frequency Response
Last week, in response to Chris Koehncke’s blog post, I set about creating a couple of sample recordings to support my belief that a headset trumps a laptop’s built-in microphone. Along the way I came to a couple of realizations, or perhaps I should say remembrances, of things that I hadn’t thought about in a long while. There are numerous subtleties to the matter of microphones.
Microphones, like most things, are built to address specific applications. There are microphones for recording studios. Microphones for stage performers. Microphones for board rooms. Microphones for mobile phones. Even a microphone for that cheesy tape recorder that you bought at Radio Shack back in the 1980’s when it was still a great gadget shop.
All of these microphones have one thing in common; they convert a sound wave in the air into an electrical signal. Beyond that they’re often very different devices. They are each designed to address their intended application.
The microphones that find their way into a laptop are not especially wonderful. Nor are they especially bad. They’re decent, inexpensive, general purpose microphones, typically an electret condenser design. They’re connected directly to the onboard audio interface (ie sound card).
More significantly, they are not in any way bandwidth limited. No effort is made to constrain their frequency response. While they don’t have the ultra-flat, wide, frequency response of a high-quality recording studio microphone, they do ok for most casual purposes.
In contrast, the headsets found in telecom are designed to meet telecom requirements. That includes bandwidth limiting the signal from the microphone. Most formal definitions of “wideband” audio in the realm of telecom are like TIA-920, which stipulates that the signal be rolled off beyond around 7 KHz.
Yes, legacy Wideband telephony standards roll-off the very high-end to protect the system as a whole. When the system is based upon 16 KHz sampling, yielding an 8 KHz pass-band.
Some systems, most notably video conference systems, support the use of Super-Wideband audio, with a useful audio channel extending up to 14-16 KHz. Production grade audio systems, as found in recording studios and broadcast facilities, aim to address Full Bandwidth audio, covering the entire range of human hearing.
In all cases, it’s necessary to filter out high-frequency detail that would be out-of-band for the system, in order to avoid nasty aliasing artifacts.
Want to know more about digital media theory, like aliasing? Check out Xiph.org’s great Digital Media Primer For Geeks. It’s awesome work by some great people, like Monty Montgomery.
So, it’s a bit unfair to compare the built-in microphones of a laptop to the microphone in even a high-quality headset. They’re not apples-and-apples. The headset, by design, rolls off the highest frequencies while the laptop does not. It makes the laptop sound a bit brighter.
From an analysis perspective, this additional high-frequency detail is easily identified using the spectral display in an audio editor like Adobe Audition. The vertical axis in the following picture (click for larger view) goes from 100 Hz to 24 KHz, reflecting the 48 KHz sample rate used to record the examples.
The laptop microphones clearly present high-frequency energy to around 12 KHz vs 8 KHz for the headset mic. Nonetheless, the headset still sounds better. High-frequency extension is not the sole consideration.
The major increase in voice intelligibility comes in the the jump from narrowband to wideband. This is why 16 KHz sampling came into widespread acceptance as what we now know as “HDVoice.”
I believe that, even when constrained to merely HDVoice, the headset still beats the laptop microphones. For when it comes to human voice, most especially humans over the age of six, there’s very little energy above 8 KHz. Optimal microphone placement, as delivered by the headset, beats sub-optimal placement by eliminating background noise and reverberation. This is true even when the more distance microphones are not bandwidth constrained.
Next Up: Noise