When silence sounds too, well, silent

Silence on a VoIP line when someone stops talking is so stark, so dead, so absolute, that people at the other end find it unnerving, compelling them to ask, “Are you still there?”

When Bob Longhini was evaluating VoIP gear, an unexpected problem came up that turned out to be nothing.

Actually, it turned out to be too much nothing. The trouble was that silence on the VoIP line when someone stopped talking was so stark, so dead, so absolute, that people at the other end found it unnerving, compelling them to ask, “Are you still there?”

The solution was to turn on comfort noise, technology that delicately blends selected sound frequencies at just the right volume to create the impression that the phone line is still alive and to satisfy the need to know that someone is listening.

“We thought, ‘Wow, some of these phones are really, really clear and quiet,'” says Longhini, who was evaluating Cisco VoIP gear two years ago for door and window maker Kolbe & Kolbe. “The discussion came up that if it’s too quiet, then people get uptight. Interesting concept: You get to the point where the call quality is almost too good.”

The eerie silence is a result of packet technology. The problem has been around for about 30 years, as long as there have been digital phone transmissions in wide use. It has been dealt with in the worlds of TDM, frame relay and ATM, and now is being addressed to make VoIP easier on the ear.

The European Telecommunications Standards Institute has scheduled tests for September to measure ear-to-ear quality of VoIP calls. Vendors will run their gear through performance trials to determine, among other things, how good their comfort noise is.

The IETF also is working on specifications for VoIP coder-decoders that include comfort noise generation, again with the goals of multi-vendor interoperability and boosting the overall quality of comfort noise.

The better these organizations do their work, the less end users notice it. Vendors are spending a lot of time, money and brain power to create something that in its best form doesn’t get noticed at all.

The inherent profound silence of packet voice technology stems from one of packets’ main attractions: When there is no voice to send, no packets get sent so bandwidth doesn’t get wasted on nothing. Because about 60% of a conversation is made up of silence, a lot of that bandwidth can be used for something else.

The downside is that if receiving phones get no packets they generate no noise, so it sounds like the connection has broken. Before packet voice technology based on TDM was developed, phone connections were analog, and the phones didn’t stop sending when speakers stopped speaking. Lines sounded open until someone hung up. Comfort noise wasn’t needed because real noise filled the void.

In packet networks, the long-standing answer to silence has been for gear at both ends to create background noise on their own. With VoIP, this comfort noise is generated in the receiving phone itself, so the noise takes up no bandwidth on transmission lines, but that is not a simple task.

Comfort noise generators and companion technology called voice activity detectors (VAD) make comfort noise happen. VAD software listens to the conversation and signals when a voice stops by sending a silence identifier (SID) frame. Silence is determined by a drop in energy, and the threshold is set on the fly, says Yann Lejas, a research and development engineer for Global IP Sound, which makes speech processing software.

An SID triggers a receiving handset to sample background noise and to imitate it as best it can. “If the comfort noise generated was too quiet or too loud or was the wrong pitch, that would bother the listener,” Lejas says. If the background noise stays constant, the next time the VAD indicates silence, no SID is sent and the comfort noise generator generates the same kind of comfort noise it did for the previous silence.

The best comfort noise doesn’t just give that open-line sound. It also tries to imitate whatever ambient noise is audible in the background at the speaker’s end. So if a person is speaking in a room with a tea kettle whistling, it will be jarring to the listener if the whistle disappears altogether when the person stops talking. It also will be jarring if the whistling is replaced by comfort noise that sounds like a vacuum cleaner rather than a tea kettle. The most convincing comfort noise is a mix of sound at the right volume that is made up of the right mix of sound frequencies to mimic what would have been heard if the call were being made on an analog line, according to Global IP Sound.

“With the processing power we have today, we can do a lot more,” says Roger Gutzwiller, marketing director of the telecom division of voice processing company Head Acoustics. Ten years ago, comfort noise was closer to generic white noise.

The algorithms used to calculate and simulate background noise are well established, but they must be adapted to fit new transport technologies such as IP. This can take time. When new technologies that can carry voice have come along, people have adopted them initially to save money on long-distance links, says Hank Lambert, an enterprise call-control product manager for Cisco. For example, voice over frame relay often was used because it could travel over existing data pipes at no extra charge and reduce long-distance costs between company sites.

For such intra-company use, comfort noise initially was left out to minimize bandwidth that voice occupied, he says. The costs savings outweighed the odd dead silence. Because all the phone users worked for the same company, they were forced to get used to it.

But as the technology matured, customers demanded comfort noise, especially if people from outside the company also used the lines. The same has been true with IP. “Initially, they decided to try it without comfort noise,” Lambert says, but soon learned that the result was too uncomfortable. Now comfort noise is pretty much a given in IP phone gear.

That is a relief to Longhini, who is evaluating VoIP for his current employer, Jennie-O Turkey Store, a billion-dollar subsidiary of Hormel based in Willmar, Minn. He’s in the midst of polling select users on their thoughts about IP voice, and it seems that with comfort noise in most VoIP gear, they have nothing to complain about. “So far it hasn’t come up,” he says.