A few weeks back, Dr. Sean Olive, longtime director of acoustic research and now senior fellow at Harman International and former president of the Audio Engineering Society, paid homage to one of the most enduring audiophile test tracks, just over 35 years after its original release, with a LinkedIn post that went something like this:
Tracy Chapman’s “Fast Car” was one of the original tracks we used to test loudspeakers at the National Research Council of Canada in 1988, and we continue to use it today at Harman. Why? Because it’s one of the most sensitive test signals to hear problems in loudspeakers and headphones.
It consistently produces the largest effect size and F-statistics in listener training and product benchmarking tests on headphones and loudspeakers, meaning that listeners can clearly hear and discriminate between the products and formulate strong preferences when using “Fast Car.”
Dr. Olive’s reminiscence and explanation goes on from there, but there’s one little tidbit missing from the LinkedIn post that he revealed in a similar posting on Facebook: “After hearing [‘Fast Car’], I introduced it into our program test signals in 1988—not knowing at the time it would become one of the staple audio test signals.”
Dr. Sean Olive
It’s one thing to have been intimately involved with the research; it’s another thing altogether to have been the one to hand-select the song that has earned such a venerated place in the history of the science of audio reproduction. As such, I couldn’t resist the urge to rope Dr. Olive into a one-on-one conversation about this enduring track and its importance in the history of hi-fi, and that tidbit seemed like an obvious place to start. What follows is a transcript of our conversation, edited for brevity and clarity.
Dennis Burger: I know you said you brought “Fast Car” in as test material without knowing at the time it would become the cornerstone of a lot of the research done at the NRC, but you still heard something in it that made it stand out to you as worthwhile test material. Or maybe I’m assuming. Did you? When you listened to the song for the first time, what was it that made you think, “I’m going to put this in the rotation”?
Dr. Sean Olive: Up to that point, Floyd Toole had been doing listening tests using mostly recordings from McGill University, which is where I studied. McGill has a sound-recording program, and Floyd had some digital recordings from there. I don’t know if they were originally digital—in fact, they probably weren’t. They were probably analog transferred to digital. He had these Panasonic VHS-tape-based digital recorders, and he used the McGill recordings because he didn’t think most mass-market recordings were very good.
With the McGill Recordings, he knew the people who engineered them, the microphones used, recording techniques, etc., and the actual monitors used in the control room—all of which helped to control the Circle of Confusion. Floyd actually designed the main monitors in the McGill control room.
But prior to the digital recordings, he was using vinyl. And one of the problems with that was that it was hard to create loops—you know, 15- or 20-second loops of music. And with vinyl, the recording itself was not stable. Over time, the record would warp or get dirty.
Digital was still new at that point, and when Tracy Chapman’s self-titled album came out, it was released on CD. When I heard it, I just thought it was well-recorded compared to a lot of other contemporary albums. It was fairly well balanced—perhaps a bit bright, but it had music that went from the lowest lows to the highest overtones, and when we played it through the speakers we were testing, you could clearly hear the differences between them. So I thought we needed to add it to our program material for testing speakers.
DB: At this point, that wasn’t for listener training, right? It was for testing designed to understand listener preferences with regard to loudspeakers?
SO: That’s right. We hadn’t even started listener training at NRC then. The only way we trained people was by running them through lots and lots of tests, and over time they developed more experience.
DB: There’s this graph you posted about the program’s effect on listening performance, where you’ve got pink noise and female pop rock at the top—and it’s my understanding that this graph shows the relevance of different listening material on determining the performance capabilities of a speaker. And then you’ve got jazz trio down at the bottom. Could you talk about what makes some material better than others as reference material? What makes “Fast Car” a better test track than a jazz trio recording?
SO: Well, the graph you’re referring to was related to the listener training. So, that was basically a task where people are asked to identify different resonances and equalizations added to the music. They have to identify, for example, whether it’s a high-shelf boost/cut, a low-shelf boost/cut, a resonance at 500Hz, a dip at 500Hz, or a peak/dip at 2kHz.
It turned out that people’s performance in correctly identifying these resonances or equalizations varied from track to track. As you said, with pink noise, they got roughly 90 percent correct, and “Fast Car” was essentially the same score. It was within a few decimal places.
So we did a third-octave spectral analysis over the length of the loop, and of course, pink noise has every frequency from 20Hz to 20kHz, and it’s very smooth on a logarithmic scale, so it has the most energy across the entire audio bandwidth of any of the tracks. And Tracy Chapman was the closest to pink noise in those respects. It has energy from nearly 20Hz to 20kHz. It doesn’t have as much energy at high frequencies as pink noise, but that’s typical of program material, which tends to have a downward-sloping spectrum.
So, yeah, “Fast Car” was closest to pink noise, and it was also very smooth. In a later study, we found that the smoothness of the spectrum is a factor. There are no sharp peaks or dips.
And the reason for why it’s sensitive is, you won’t hear resonances, bandwidth limitations, etc., unless there’s a signal that excites them. So if you’re just listening to female speech, for example, you’re not going to know if there are any problems below 100 or 200Hz, because there’s no such information in the signal. Similarly, above 10 or 12kHz, you may not hear whether the speaker is rolling off.
Floyd Toole and I also did a study on the detection of resonances where we adjusted the absolute level or the threshold at which you can hear it, and it turns out that pink noise was the signal where you can hear the lowest-level peak or dip in a signal, so for a Q of, for example, 1, we can detect differences of half a dB or less. And if you’re listening to music, it’s on the order of ±1.5–3dB, so we’re much more sensitive to broadband continuous signals than most music, but Tracy Chapman is, again, closest to pink noise in that regard.
DB: When did you start to dig into those considerations? In other words, when did you start to put some serious thought into why this was such a good test track?
SO: We realized it when we did a statistical analysis of the listening-test results. People would come in and do speaker comparisons, and you always look at how the scores of products vary with each different program. And you hope the program is not a factor at all if you’ve carefully chosen your music.
But with “Fast Car,” people were much more discriminating. They were better able to form preferences among the speakers with this test track. With other test tracks, you’d put it on and people had more trouble deciding which speaker they preferred—or, later on, which headphones they preferred. So we had the statistical evidence that people are better able to do the tests—they come up with answers much faster and much more consistently and with much more discrimination. That makes a great test signal.
DB: If you were having a conversation with someone who’s just getting into hi-fi, and they were looking to maybe go to a stereo shop or two and audition some gear with “Fast Car” in-hand, what would you tell them to listen for? What would they be hearing that might tell them what makes one speaker better than another?
SO: First, listen to the bass. There’s a fair amount of it in “Fast Car.” The song has an electric bass and kick drum. Listen for which speaker can recreate the depths of bass better. Secondly, the hi-hat is quite prominent in the mix. Is it sparkly? Is it dull?
Also, listen carefully to her voice. This track has been found to be very revealing of distortion. The combination of the bass and her voice is such that if the speaker has trouble reproducing the bass—and we’ve heard this with bookshelf speakers—it’ll basically max out the woofer excursion, and then it will start modulating her voice.
You can actually hear her voice wavering a bit in pitch. This is something that Wolfgang Klippel, who makes test and measurement equipment, became aware of, and when he does seminars that demonstrate how to measure distortion and nonlinearities in speakers, he always uses “Fast Car” because it’s so revealing.
I think one of the reasons is, not only is the bass prominent, but there aren’t a lot of things around it, so if there are any distortion components, they’re not well-masked by other information in the recording. So the distortion tends to be quite audible.
Home of the world-famous NRC anechoic chamber
We’ve been playing this track for 35 years, and people often complain that it’s so old. “Why play it? I’ve heard this a million times.” Some people say it’s not a very good recording. And it’s certainly not a perfect recording. It’s a little bright, as I mentioned before. But as you become familiar with the track, you accommodate for that brightness, to the point where if I hear a speaker that’s not a little bright with “Fast Car,” then I know that speaker is dull.
The other thing about recordings that have problems—when you’re comparing them across multiple comparisons, across different speakers or headphones, is that the faults tend to be common between all speakers, unless the fault has a complementary character. If you have a peak in the recording at 10kHz and you have a dip in the speaker or headphone at 10kHz, they may cancel each other out. But by and large, the faults with recordings tend to be common across all speakers. Your brain tends to just ignore these problems, and you focus on the things that are different between the speakers.
The original NRC listening room for blind testing of speakers
DB: Let’s say you’re sitting down to listen to music on that amazing system I see over your shoulders there, with no agenda. Could you put on “Fast Car” and simply enjoy it for what it is, without picking it apart or analyzing it?
SO: It’s funny—I rarely listen to it for pleasure. Probably because it seems too much like work [laughs]. You know, when using the song as a test signal, we only used the first verse, and sometimes I’ll be in China or somewhere in the world, and I’ll be in an elevator or hotel lobby, and the track will come on and I’ll realize I’m completely unfamiliar with the second verse.
So yeah, it’s not a track that I routinely sit at home and listen to.
DB: Switching gears completely here: I know you have a no-country-music policy when it comes to test tracks. What’s that about?
SO: That’s mostly a personal preference. Mind you, country music has changed a lot over the decades. It’s more of a pop style now. But it’s such a combination of various styles that it’s probably not fair to keep it out of our playlists.
DB: Growing up in Alabama, I was subjected to so much country music that I hate it now. But I make three exceptions: I love Lyle Lovett, Willie Nelson, and Waylon Jennings.
SO: Oh! I love Lyle Lovett, too. Joshua Judges Ruth was a very well-recorded album. I think it was recorded by George Massenburg.
DB: I use cuts from that album as reference material in every gear review I do.
SO: And I used one of the tracks in my PhD research. There are some amazingly wide-bandwidth tracks in there. Really excellent recording.
DB: And Willie Nelson is one of the best singer-songwriters in the history of popular music. So maybe make an exception for Willie.
SO: [Laughs.] Fair enough.
. . . Dennis Burger