Evaluating USB Headsets for Interpreters

This is admittedly a deep dive into a niche topic. It stems from work done for ZipDX, but is more technical than most audiences can stand. Nonetheless, those of you who frequent these waters may find it interesting.

Why Do This?

One of the more fascinating aspects of my work at ZipDX involves the interpreters engaged in the use of our multilingual conference capability. These people, who are located all over the globe, are simply fascinating people. They have incredible skills with languages, and finely tuned sensitivity to the nuance of cross-cultural communication. It’s positively inspirational to hear them at work, and very gratifying to support them in their work.

Simultaneous interpretation (SI) involving an audience of more than one person by definition involves some kind of audio equipment. In facilities setup to handle such conferences there are interpretation booths that house specialized audio equipment. Such installations are costly, but the task is critical to the organization, so the cost is justified. Such facilities invariably have a technical support staff to keep it all working correctly. So, the role of the interpreter is to interpret. Beyond the basics of its use, they’re not responsible for the technology.

Things are quite different when an interpreter decides to accept assignments delivering simultaneous interpretation over-the-phone using ZipDX Multilingual. A ZipDX multilingual call is a virtual meeting, the participants, including the interpreters, can be scattered around the globe. In such a circumstance the interpreters must take responsibility for the tools that enable them to work remotely.

When an interpreter is preparing to use ZipDX for the first time we ask them to go through a short self-starting guide before we schedule live training. The Getting Started Guide describes the technical requirements for the work, including the requirement for a headset.

An interpreter may already own a headset designed to connect to the console in a traditional interpreter’s booth. It may or may not be useful with respect to their computer. Further, at ZipDX we are quite demanding of headsets. We want the interpreter to hear the best sound possible, since that makes their job easier. We also want the audience to hear the best sound possible. Careful attention to delivering quality audio ensures that everyone involved has a great experience.

As a consequence, we occasionally get asked to recommend headsets for use by interpreters who are starting to deliver SI over-the-phone.

Keeping Things Simple

When making recommendations to interpreters we prefer USB-attached headsets. The logic of this preference is simple;

1. There’s only one kind of USB connector used for headsets.
2. When a USB headset is connected to a computer it’s almost always clearly identified by the operating system.

Of course, most computers have built-in audio capability. People often refer to this as the “sound card” even though it hasn’t been a separate card for years. Someone can use an analog headset connected to their computer and achieve acceptable results. This is very simplest approach since there only one audio device available, meaning that you can’t select the wrong one.

However, there are a range of different connector schemes used for analog headsets, that may or may not match the computer in question. An adapter may be required, which is just something to forget, or worse, to lose.

Further, the audio processing in a USB headset is designed for voice applications. Better models have on-board signal processing (DSP) designed to eliminate noise and echo. The same cannot be said for a typical sound card.

There’s a catch: Double-talk

While we prefer USB headsets, we have found that not all such devices are created equal. Some suffer problems coping with double-talk. Let’s begin by understanding what this is and why it matters.

While we may be in throes of the US election primary season, this kind of double-talk is unrelated. In this case double-talk is the term used to describe when an audio device is both sending and receiving sound at the same time.

In a normal, polite conversation double-talk is not usually a factor because, except in rare cases, we don’t talk over each other. One person speaks, then the other responds. While there may be some talk-over, it’s not a free-for-all unless someone’s especially argumentative.

The working life of a conference interpreter is quite different. They are always listening to the person speaking on the floor, even as they repeat what they hear in another language. Their headset is continuously experiencing double-talk.

What we have discovered is that some USB audio chips used in headsets degrade the sound from the microphone while there is sound playing in the earpieces. Since the effected sound is outbound, there interpreter cannot hear this. The audience hears it as annoying changes in the sound of the interpreter.

A detailed investigation has revealed that the sound from the interpreters microphone is actually toggling between wideband and narrowband in reaction to the level of the sound present in the earpieces. This is very plainly a fault in the design of the USB audio chip.

When we first discovered the problem we reported it to Plantronics, the manufacturer of the headset in question. We described the problem in detail, including documenting our test setup and providing samples of faulty audio.

They launched their own internal investigation, then went quiet for a couple of months. They eventually reported that they had observed the same behavior. They remarked that it was not something that could be remedied in-the-field, and offered to replace the headset in question with a better model that did not exhibit the problem.

We accept that ours is a niche application. Most people do not require perfect audio in both directions at the same time. Most people never even notice the fault that we had identified. Plantronics’ response was about all we could reasonably expect.

In the period since this fault was initially discovered we’ve testing a number of USB headsets. We’ve created a list of headsets that we can recommend to interpreters, where each was known to be well-behaved. Along the way we’ve found a number of headsets, from several leading manufacturers, that exhibit this very same fault.

Testing Double-talk Performance

The test process I developed works as follows:

• Use a media player application to play a bit of reference speech to a loudspeaker
• Use a second media player to play reference recording to the headset
• Position the headset so that the loudspeaker mimics a user
• Record the output of the headset microphone for evaluation

I’ve now staged this test enough times that I can do it very quickly using VoiceMeeter Banana, two instances of VLC, a ClearOne Chat 160 and the headset under test.

In one instance of VLC I play a reference recording (a story from NPR’s Marketplace) to the Chat160. In the second instance of VLC I route a recording of my voice reading from a book. In VoiceMeeter Banana I record the headset microphone to a WAV file (Mono, 16 bits @ 48 KHz.)

Evaluating The Results

To evaluate the recordings I load them into Adobe Audition and look at the energy vs frequency view. Each if these images is linked to a high-resolution original screen capture.

The Reference Recording Of My Voice

The Y-axis denotes frequency, which goes from 200 Hz to 8 KHz, consistent with the best traditions of wideband telephony, which is all that such headsets can deliver. The white line across the middle is at 4 KHz, just above the limit of narrowband telephony a la PSTN. Note that there is consistently energy present all the way to 8 KHz.

Headset #1: Microphone Audio with No Sound To The Earpieces

The headset under test in this case is a VXi Passport 21V with a VXi X100 USB adapter.

This first test involved examining the sounds from the microphone while there’s no sound being played to the earpieces. Notice that the image looks nice and consistent, very much like the reference recording. The microphone is delivering clean, wideband sound.

Headset #1: Microphone Audio with Sound Playing To The Earpieces

The headset under test in this case is a VXi Passport 21V with a VXi X100 USB adapter.

This image shows that the sound from the microphone remains consistently in wideband, even when sound is playing to the earpieces. This particular headset does not exhibit the fault.

Headset #2: Microphone Audio with No Sound To The Earpieces

This test is a different USB headset. Unlike the earlier model, this one has the USB interface permanently built-in.

With no audio playing into the earpieces the stream from the microphone remains clean and consistently in wideband.

Headset #2: Microphone Audio with Sound Playing To The Earpieces

This test is a different USB headset. Unlike the earlier model, this one has the USB interface permanently built-in.

With audio now playing into the earpieces it’s plain to see that there are portions of the microphone stream that fallback to narrowband (the white line at 4 KHz.) This headset exhibits the fault describe previously and cannot be recommended for use by interpreters.

To make it painfully obvious, the above image presents a comparative view. The clean audio is on the left and the degraded version on the right. Click on the image to see a full resolution version of you like.

Summary

When the fault is observed I report it back to the manufacturer. I’m happy to provide them with supporting evidence. That’s what made it practical to create this very blog post.

The fault may well be linked to a particular piece of silicon that’s in widespread use. Even if we could identify the chip in question it wouldn’t help, since at the OS/application level each headset is identified by it’s make & model, not the ID of the chip inside.

If the problem is in fact a fault in silicon then there’s nothing to be done about it except make note of it, and not recommend that product. Hopefully, with a newfound awareness of the issue, manufacturers will be more careful in future and not use the same chip again.

You’d think that common USB audio interfaces would be a mature, stable technology, without much variation between implementations. I’ve discovered that’s simply not the case. I’ve now tested a dozen USB headsets and found about a third suffer this problem.

The early results of this effort have appeared in a blog post by Barry Olsen GM of Multilingual at ZipDX and Co-Founder of Interpret America. More recently we’ve published a list of recommended headsets at ZipDX.info. Our further intention is to offer a series of reviews over time. We prefer to recommend known good products, as opposed to flagging those that exhibit this  problem.

P.S. –  I’d like to thank VXi for their cooperation in my exploration of this issue. Of the various people that I’ve dealt with VXi has been the most helpful and willing to take on-board my concerns.