Meeting Topic: The Perception and Measurement of Headphone Sound Quality
Moderator Name: Sean Olive
Speaker Name: AES, Harman International
Other business or activities at the meeting: Happy Holidays
Meeting Location: Ivar Theater, Hollywood, CA
AES-LA gathered at the Ivar Theater in Hollywood on Thursday, November 21st, to hear about Sean Olive's headphone research at Harman International.
Sean Olive summarized a 7-year research project aimed at understanding the perception and measurement of headphone sound quality. There is more research to be done, but the team is moving to more pressing things. The research team included Sean Olive, Tod Welti and Omid Khonsaripour.
Jeroen Breebaart published a research paper in 2017 showing no correlation between headphone frequency response and retail price over 283 headphones. (Journal of the Acoustical Society of America) Clearly, the industry was in need of scientific guidance in how to design and test headphones for best sound.
Testing headphones presents unique challenges. You can put speakers behind an acoustically transparent curtain so people can't see what they're listening to, but you can't do that with headphones. Blind tests were problematic because the test subjects would identify the units by weight and relative comfort.
To achieve spectral balance, the frequency response must have a peak at around 3kHz, the resonant frequency of the human ear canal. Many headphones do not have this necessary peak. Some closed back headphones don't seal well against the head and the leaks result in weak response under 500 Hz.
Do trained listeners agree on what makes a headphone sound good? Six popular headphones ranging from $115 to $995 in price were placed on the heads of trained listeners. The intent was for the test to be blind, but the test subjects began to identify the headphones by weight and comfort, so another method was developed to perform the tests. Still, there was enough data to show some agreement among trained listeners. Preference ratings began to correlate with frequency response, measured using a GRAS 45CA ear simulator.
Virtual headphones: To focus on identifying a target frequency response and remove visual, brand, price, weight biases in listening tests, equalization was applied to simulate the response of various headphones.
Current standards in 2013 in the headphone industry specified target curves in diffuse-field and free-field curves (ISO 11904-2). It became obvious that few manufacturers were following the published standards based on the wild variation of measured frequency response, possibly because of not being aware of the standards or possibly because of not believing in the standards.
"The hypothesis was that a headphone's target response should ideally approximate an accurate loudspeaker in a listening room," Sean said.
A mannequin with pinnae and ear couplers was placed in the Harman reference listening room in three head positions. Thusly, the team measured the steady state response of the loudspeakers. This provided a new target curve.
Validation of Headphone Target Response: The following procedure was used to correlate objective measurements with subjective listeners' impressions and validate the headphone Target Response.
Listening tests were done using two different headphones: Sennheiser HD518 and Audeze LCD2. Equalization was applied to make these units perform with various published target curves including the new "Harman Target Curve" and a second one that has less treble and bass. Each target curve was rated by trained listeners based on preference using three music programs with one repeat. The Harman Target Curve was widely preferred over all other curves including no equalization using both test headsets.
The team did tests in Canada, US, China and Germany using 238 both trained and untrained listeners comparing 4 virtual headphones. There was good correlation between the preferences of trained and untrained listeners. The preferences were the same among all nationalities. People generally like an increase in bass, smooth response without sharp resonances. The best-selling headphone came in last. It has lots of extra bass, but also has a big hole at 1 kHz, so it's sucking out the midrange making it sound dark, closed and colored.
The team turned their focus to in-ear headphones. Do in-ear headphones require a different target curve? It was evident that performance below 500Hz depends on a good seal with the ear canal. Test subjects were given controls over equalization and level. A new target curve emerged with approximately 10dB more bass at and below 125 Hz than for the around-ear target curve. Fifteen commercially available in-ear headphones were tested. Not many come close to the new in-ear target curve. Most gave extra bass from 125 Hz to 500 Hz with additional deviations at and above 3kHz. There is plenty of opportunity for improvement.
Tests were done using a virtual headphone technique replicating the frequency response of in-ear headphones. Thirty different models varying in price from $26 to $1000 were tested using 36 trained and 35 untrained listeners in California and Michigan with a median age of 35 years, 82% male and 18% female, all with hearing tested as normal. The new IE target curve was preferred over all other curves. The results were similar for trained and untrained listeners. The Samsung Galaxy Buds are designed to follow this curve.
More recently, cluster analysis of headphone listening test data has shown there are three segments or classes of listeners based on similarities in their headphone sound preferences. Both demographic (i.e. age, gender, listening experience) and acoustic factors are associated with membership in each headphone segment. This information can help guide future headphone design that is aimed at a specific class or segment of listener.
The research has resulted in 19 published papers, two book chapters and a standard measurement technique which predicts the sound quality rating and is being used by several measurement companies.
Video of this presentation can be found at vimeo.com/lafs. Scroll down to find the title slide.
Sean's Audio Musings blog can be found at SeanOlive.blogspot.com.
www.audiotestkitchen.com is a website allowing you to compare microphones using recordings done in Harman's lab.
Sean Olive is Senior Fellow, Acoustic Research for Harman International, a major manufacturer of audio products for consumer, professional and automotive spaces. He directs the Harman X Sound Quality Group, and oversees the subjective evaluation of new audio products. Prior to 1993, he was a research scientist at the National Research Council of Canada where his research focused on the perception and measurement of loudspeakers, listening rooms, and microphones. Sean received a Bachelors in Music from the University of Toronto, and his Masters and Ph.D. degrees in Sound Recording from McGill University in Montreal. He has written over 50 research papers on perception and measurement of audio for which he was awarded the Audio Engineering Society (AES) Fellowship Award in 1996, two Publication Awards (1990 and 1995) and best peer-reviewed paper award at AES 143rd in 2017. In 2013 he was awarded the ALMA Titanium Driver Award for scientific contributions to the loudspeaker and headphone industry. In 2019, he received the Harman External Leadership Award. Sean is an AES past President.
Thanks to Sean Olive for sharing his research. And thanks to AKG for providing the headphone door prizes.
Written By: Tom Levno