I recently recorded a solo classical pianist using the audio software, simple multitrack, that I have been writing. We recorded at a medium-sized auditorium in Westchester. So far, we have had two sessions and there will probably be more.

In the first session, we had all kinds of technical difficulties. Most of these were related to bad sound quality. I brought a decent microphone and a laptop and a small mixing board because the microphone needs phantom power. There was a buzz in the input that I struggled to get rid of. Although it turned out that we had access to the auditorium's Mackie mixing board, which is better than my ancient mixer, that didn't help. It seemed that the laptop was quiet on its own, and the mic and mixer were quiet on their own, but when the mixer was plugged into the laptop the noise would appear.

I used to work for a sound company. In live sound, one of the standard techniques to combat noise problems is to put an isolator between two devices. An isolator breaks the ground loop. Only the differential signal on the balanced line passes. I improvised an isolator from a little transformer, and it did improve the signal-to-noise ratio (SNR). But, as expected, it was not passing the full range of audio frequencies, and so it sounded like an AM radio. Good isolators exist but I didn't have one.

One of the principal frustrations with trying to get a good SNR in recording is that it is not a quantity that can be measured simply. There are five main things that one has to keep in mind.
  • overload of the analog to digital converters (ADC) of the computer
  • overload of the microphone
  • overload of the analog signal path
  • quantization noise in the ADC
  • spurious noise from radio/powerline interference
    These concerns are somewhat conflicting, and so to get a really good recording one has to balance them. The frustration is that most of these things are difficult to observe directly. For example, when something is overloading and it sounds raspy, you can't (or anyway, I can't) tell exactly what is overloading. So it involves experimentation. But there are many dimensions so the search for great sound can be slow.

    Eventually we did find decent sound, and we left the session satisfied. But I made two conclusions. First, it is desirable to have my own excellent mixer that interfaces well with the computer's audio input. Second, it is desirable to somehow quantify the variables of the recording process so that it isn't so mysterious and daunting.

    For the purposes of recording piano, one microphone can be enough. So a mixer isn't necessary, just a really good microphone preamp. It needs to provide phantom power. I decided to build the best preamp I could for this purpose. There are lots of schematics on the Internet. They are mostly the same. I tried to see how the mic inputs are implemented in the really respected audio equipment like Neve and SSL. It turns out that the high-end stuff often uses transformers. But the transformers are custom/secret designs or very expensive off-the-shelf parts. It seems that most of the newest designs are being done without transformers. It isn't clear whether this is because of technology improvements or because of the market pressures to make things cheaper. Most of the new designs, even good stuff, uses audio opamps, no transformers, no discrete transistors.

    Regular opamps aren't very good for mic preamps. To get good audio performance, it is necessary that the input transistors of the opamp are biased with a high current. Most opamps try to minimize the current they use, and so they aren't good for mic preamps. Two companies that make excellent audio opamps are Analog Devices and Texas Instruments (Burr Brown).

    I built a preamp from an Analog Devices AD797. The circuit is nothing special, just pretty much the same as what everybody does. The voltage gain is 100. I used to work at a chip company that made parts for the communication business. It is common knowledge in that field that thing first amplifier in the signal chain is usually the one that determines the noise figure for the whole system. But that is only true if the first amplifier has significant gain. A gain of 100 is usually sufficient to make the noise of the other stages negligible. The amplifier works well and it was used on the next recording session, as explained below.

    Before the second piano recording session, I wrote a little program called det to help make recording more quantitaive and less mysterious. It is meant to be used with my simple multitrack software. It gives a scrolling peak level indicator and prints warnings about any audio samples that reach the maximum level, clipping. Det quantifies the ADC overload situation, and informs the user of how much headroom there is, but there was an unintended benefit. The signal to noise ratio (SNR) becomes quantifiable. Det prints the actual peak sample value during each 500msec period. While recording silence, this can be used to measure the noise floor of the recording.

    Ideally, the peak values during the recording would be almost clipping, but not quite. The clipping point is around 32767. And during the silence, the sample values should be zero. The values don't go down to zero. They do down to some finite number, like 70. Since it is now possible to measure the SNR, it is possible to work systematically at improving it. For example, it is possible to get an idea of how much noise is coming from each part of the signal chain. It is also possible to compare one microphone preamp with another. I confirmed that the SNR that we were getting with the home made mic preamp is better than the best we could do with the Mackie mixer that belongs to the auditorium.

    Mic placement is a lot of fun. It is very subtle but makes a huge difference in the recording. There are two purely technical things that need to be avoided. If the mic is too close to the piano and overloads, it will sound horrible. If the mic is too far away, the noise of the mic and mic preamp will become a problem. But these are extremes. The distance of the mic from the piano determines how ambient the recording is. The hall that we were in had good acoustics, so when we tried placing the mic a little bit further from the piano, the sound became very luxurious, but it lost its focus. When it was a bit too close the sound became too clinical, not sensual. Again, it is a balance. There are people in the world who are experts in mic placement. I'm sure that they have their secrets. I went to a big bookstore and simply read everything they had on the subject of piano miking techniques. We tried several things that the books recommended and several crazy experiments. The conventional positions described in the books seemed to work best.