Audio Quality Matters

sample rate of 44.1 thousand samples per second (expressed as kilohertz per second or kHz) and a 16-bit sample size. Samples per second defines the frequency range of the transmission.

In digital encoding of an analog signal, the sample rate must be slightly higher than double the effective frequency range being converted to digital. This is known as the Nyquist ratio and says that two samples must be used to determine the state of the signal at that point. The CD uses 44.1 kHz as the sample rate with the half value of 22 kHz so that a filter to remove the noise of the sampling  results in a frequency range up to about 19-20 kHz, generally considered the limit of human hearing. The range of the human voice extends from 80 Hz to 14 kHz; however, the 8 kHz sampling of the traditional digital phone limits audio frequencies to the range of 300 Hz to 3.4 kHz. This challenge is further exacerbated by the relatively low dynamic range that an 8-bit sample results in. In fact, a phone call has only 255 levels of volume, while the CD has over 64,000. For music, even the CD is no longer considered adequate, as modern music mixers generally use a 192 kHz sample rate and 24-bit samples. The reason for all of this is quality. Increasing the sample frequency increases the frequency spectrum; increasing the sample size enables a much better signal-to-noise ratio and much better clarity between intonations. The combination produces dramatically better quality. The inverse is equally true, as compression used in legacy wireless technology is even worse in both size and quality, “Can you hear me now?”

So the phone system is compromised due to the time period and technology limitations of development. The devices that have traditionally been used for telephony are defined by these limitations as well. There is little reason to have wider speaker frequency response when the actual transmission is so limited, so traditionally, manufacturers have chosen components that reflect the low frequency response. This limitation of design has been extended to include the relatively low dynamic range of the 8-bit samples. In conference devices, simple microphones and speakers have been adequate; they can deliver quality equivalent to the transport network.

 

To clearly understand how limited the phone system is, listen to any wideband audio content over a traditional phone. Listening to music with a full frequency spectrum and high dynamic range demonstrates

how completely the experience is masked by the poor quality of the system. In fact, there is a reason that there are specialized companies selling music on hold content; it is content that is optimized to the relatively poor audio characteristics of the current telephony network.
 

The reasons for this are multi-fold. The low frequency limits some parts of the human voice, the resonant frequencies that make up our “sound.” While early developers of telephony rightfully noted that the PRIMARY frequencies of human speech fall in the 300 to 3.4 kHz range, their antiquated testing equipment do not tell the whole story. First, the speaking voice of a typical male adult willhave a fundamental frequency from 85 to 180 Hz; a typical female adult from 165 to 255 Hz. So, in a typical conversation, much of the speech may actually be below the G.711 range. Our brains can interpret the higher frequency harmonics generated to actually hear what is said, however, this makes the sound nasally and harder to understand. Similarly, the higher frequency harmonics may be dropped, resulting in a flat sound.
 

Figure 1 shows an actual frequency spectrum of a human male speaking as recorded over a one-minute period, showing the relative level and frequency of the audio generated. Note that while the predominant frequencies are in the 100 Hz to 1 kHz range, there is significant content beyond that range. The reality is that the overall sound of the human voice is the sum of all of these frequencies. The reason for this is a factor of both sound and human anatomies. When we talk, we generate a basic frequency, but certain sounds make much higher frequencies. For example, the hard “s” sound comes with a frequency that is much higher than the typical range. Naturally these are small variants, but they are well within the normal range of human hearing and are critical to fully understanding the spoken word. In human speech, this is further exasperated by the natural resonance of the human voice box and the mouth that “tunes” the speech. In addition, having the dynamic range to make these frequencies audible versus the fundamental is critical. The result is that while the fundamental frequencies of human speech are relative low and can be defined to fall within the legacy telephony spectrum, the full spectrum has a range of frequencies at a wide range of dynamic levels that combine for the full speaking experience. The implementations of the telephony system have left much of these extended values on the ground, resulting in a perception of telephony as flat

and “lifeless.”

The movement of telephony voice from the simple 64Kbps basic telephony to a higher resolution and more dynamic audio has been caused by VoIP and the capability of using compression as well as higher bandwidth to transport the audio traffic. In VoIP, the voice can be carried as G.711; however, even the earliest VoIP solutions enabled compression. In the early days of VoIP, this was done to reduce the bandwidth. However, as bandwidths on all networks continue to increase, the percentage of traffic that is generated by voice is decreasing. Networks tend to have a 10x increase in bandwidth every 7-10 years. Whether desktop, nomadic, or wireless, this increase in bandwidth is driven by web, applications, and predominately in recent years, video. Clearly the number of voice calls is not increasing at this rate, so the possibility of using the available bandwidth for higher quality emerged.

The first step in higher quality voice was using compression techniques to put a higher quality voice signal into the same network bandwidth. The first attempts at HD voice used the G.711 codec technology with a higher sampling rate. The G.722 codec is generally considered the baseline for HD voice; it captures and delivers sound between 30 Hz and 7,000 Hz. While the G.722 is essentially double the frequency of a G.711 codec, an HD voice call using G.722 consumes the same 64 Kbps of bandwidth as the digital POTS equivalent of G.711. So moving from G.711 in a VoIP phone system to HD voice can be done without needing more bandwidth.

While this was the start of HD back in the early 90s and was supported by a range of companies, the introduction of web-based solutions has accelerated the quality that a VoIP session can carry. For example, the SILK codec used in Skype can go up to 24 kHz sample rate, yielding an effective frequency band up to over 10 kHz. It does this with a bandwidth of around 40 Kbps plus packet overhead. Figure 2 shows the definitions of different types of codecs for audio as defined by the Internet Engineering Task Force (IETF).



The latest advances are in codecs like the OPUS codec. OPUS is an Open Source codec used in web technologies like WebRTC. OPUS is designed for both fixed and variable bit rates (where the rate changes based on the content). OPUS can operate at a sample rate of up to 48 kHz, effectively delivering a codec that can cover the full range of human hearing. In fact, audio codecs used in both telephony and video over IP have a full range of capabilities. Figure 3 shows the range of codec options available today, from very low bandwidth to full range including stereo. The MP# and AAC codecs are typically used in audio for either recording or streaming of personal audio. While this chart shows frequency range, the dynamic range is also increasing, with OPUS having a selectable sample size of up to 24 bits.

The result of all of these changes, driven by inexpensive compute power, ever-increasing bandwidth and efficient low cost open source codecs is that more and more of our “telephone calls” resemble a high quality audio connection. However, as the components of the audio chain have remained unaltered in many cases, the actual audio is impacted by the devices. Thanks to digital technology,the microprocessor and software, what was once the strength of the system has now become the weak link.

For the layman, the difference between a high quality audio stream and a low quality one is often challenging, however the difference between legacy telephony audio and the modern high definition audio for IP-based collaboration is clear. In normal speech, a quality audio codec and implementation will make the human voice more understandable. This is because the system will reproduce both the full frequency range as well as the dynamic range that today’s communication and collaboration systems can deliver.

For example, when a traditional G.711 call is played through a high quality speaker, the sound will be flat and understanding some speech points may be challenging. Playing the same call using an HD codec will result in a warmer and easier-to-understand conversation. However, play the same HD quality conversation through a low cost USB speaker and it will reflect the limited quality of the source. The challenge for modern communications speaker developers is to deliver a full range audio signal, something that is often achieved with multiple speakers in the high fidelity audio world. Limiting a device to a single small speaker and expecting intelligible sound is allowing the last challenge in the new world of audio to persist.

While the audio codec, bandwidth, frequency and sampling rate define the audio signal, the environment where the user and the device exist can be more important. In a small space, while a small speaker may work, it will not have the presence to get the full nuance of speech. In a larger space the challenge is generating enough volume to fill the space without over-driving the amplification or speakers, resulting in distortion that can hide the subtleties in speech.

In small and large spaces, an external microphone can be an issue. This can show up in many ways. For example, in a small room there may be echoes in the room itself that impact hearing what is said. In a conference setting, the device used generally should have multiple microphones as well as adaptive circuitry to select not only the best microphone, but also to enable gain matching so users do not sound like they are at the end of a pipe. Managing these multiple microphones is a challenge, but also one that is amplified by the need for excellent echo cancellation.
 

One final key to a great audio conversation is a lack of echo. While echoes can be introduced in the legacy system, they often are generated by end points in modern VoIP systems. For example, many PCs have variable timing between the microphone and the speaker. If the application is doing echo cancellation (removing the input from the output), the large time domain and variability of the audio through the internal components can result in echoes getting back into the conversation. While these may originate at one device, the resulting echo is heard at the other end, or on a conference call, by everyone else. In the office environment a headset is

a critical tool to manage echo as the separation of microphone and speaker naturally reduces echo. However, in a conference room or using a speakerphone-type device in an office, the key to echo cancellation is managing the audio environment, something that a great conference device can do. By using multiple microphones combined with active tuning and echo cancellation, a great audio device can sound as good, or even better than a headset.

In the new world of high definition audio, combined with the bandwidth and power of the Internet, a new age of voice and audio communications is exploding. The advent of the Real Time Web and multiple new applications

using VoIP HD audio drive an ever-expanding need for higher quality voice audio. By combining the full frequency spectrum with higher dynamic range resolution and audio devices specifically designed for this new environment, conversations and conferences can be much easier to understand and participate in. The resolution of these new systems enables users to understand small variations and more readily identify speakers from their voice patterns, as well as understand the nuances of human speech, that if not noticed, might result in misunderstandings.
 

As we all go on this journey to the new advanced capabilities of a high definition voice conferencing and collaboration world, making sure that all elements of the solution are aligned to optimize the overall experience

and assure understanding is critical. The selection of the device to put in a conference space, whether in a dedicated conference room or in an office will have a major impact on the collaborative events that are conducted in the space. Choosing devices with the right speaker and microphone solutions that are appropriate for all your conferencing spaces will help ensure you are getting the most value from your new communications and collaboration solution.