Simplify voice processing for home network systems

Interactive applications operated by voice recognition, for example integrating door security systems and the ability to control home appliances, are key features of home automation networks. This interactive capability depends on high-quality voice processing technology, including acoustic echo cancellation, low signal distortion and noise reduction techniques. A home automation system must also be scalable to allow future evolution, flexible to support field upgrades, interactive, easy-to-use, cost-efficient and reliable.

This article introduces some of the voice quality performance issues and design challenges unique to home automation systems. It will discuss home automation network applications that rely on voice processing, and examine some of the critical features and functionality that can help ease design complexity and cost to deliver enhanced performance.

Voice processing in home networks
The home automation market is moving beyond high-end luxury homes to target the mainstream consumer. Even in its infancy, researchers estimate the market is worth over $1 billion. In Asia, Europe and North America the home automation market is growing at an average of 10% per year. In Europe alone, demand for home automation systems is expected to double by 2009 to create a $0.5 billion market. While the market grows, home automation systems themselves are evolving to incorporate technologies such as Bluetooth, Wi-Fi, X10, ZigBee and TCP/IP.

As the market and technology matures, high-quality voice processing performance becomes increasingly important for both automation and security applications. Voice is an enabling technology that unifies the home network and is used to control appliances, telecommunication, security and entertainment equipment. End users are more comfortable communicating with a human voice, rather than interacting with a machine.

Poor acoustic echo cancellation, ambient noise and signal distortion make it increasingly difficult for a home automation system to perform reliably. If impended by poor voice performance, voice recognition cannot easily detect commands to turn on/off the appliances and voice authentication has difficulty verifying the user to allow access to the residence.

Integration of telephony and intercom
Home automation systems are increasingly integrating telephony communication and intercom functionality. In some system designs, the intercom panel becomes the main communication tool and serves as both the residence access monitor and a hands-free telephone. Therefore, the quality of voice communication becomes an important aspect of the overall system.

An important feature of a high-quality speakerphone is full-duplex operation and good double-talk performance. A full-duplex system allows two-way voice signals to pass simultaneously, enabling a natural conversation. A system with half-duplex fallback will only allow the signal with the highest power level to pass and cuts off the other signal during double-talk. As a result only one side of the conversation can be heard and the conversation becomes choppy and unnatural.

In a full-duplex system, good acoustic echo cancellation is achieved when the algorithm continuously converges, even during double-talk situations. When the algorithm continuously converges it tracks changes in the echo path as the signal source moves. If the algorithm stops to converge during double-talk and resumes convergence when double-talk ceases, users will hear an audible burst of echo while the algorithm re-adapts to the new echo environment.

For digital speakerphones, another aspect of achieving a good quality call is the choice of codecs (coders/decoders) that perform the signal conversion from digital to analog and vice-versa. Traditional mid-quality digital phones use narrowband codecs and a sampling frequency of 8KHz. Wideband codecs enable higher speech quality by doubling the sampling frequency of narrowband codecs (16KHz vs. 8KHz) and capture more components of the human voice to enhance speech quality.

Designing a high performance system
Smaller intercom panels are now being designed to be less obtrusive as part of an overall home architecture. The restricted panel size creates limitations on the design of the plastic enclosure, the choice of microphones, speakers, their placement and separation.

In an ideal system the separation between speakers and microphones should be maximized to optimize the performance. In a small plastic enclosure, the reduced separation between the microphone and the speaker leads to excessive acoustic coupling and results in louder echo. Designers are faced with the challenge of selecting smaller speakers to fit the limited space within the plastic enclosure while having to meet high volume audio requirements. The designer is then forced to drive a small speaker into a non-linear range. This results in an increase of total harmonic distortion in the plastic enclosure.

To solve this problem the designer needs a voice processing solution that can handle more distortion in the echo path to compensate for poor speaker performance. An algorithm that can cancel linear and non-linear echo allows the designer to drive the speaker volume higher while minimizing distortion. The non-linear echo canceller compares the residual echo signal " from the linear echo canceller " with the reference signal and subtracts the estimated frequency components.

Figure 1 shows a block diagram of linear and non-linear echo cancellation. This circuit removes linear echo and non-linear distortion components. It allows the system to overcome poor acoustic echo cancellation caused by distortion in the echo path.