The ever-evolving technology landscape, dominated today by cloud migration, omnichannel communications, digitization, speech, artificial intelligence, and voice bots, is reshaping the way businesses engage with their customers. This transformation, coupled with the increasing emphasis on automation and self-service over agent-assisted calls, has made the quest for delivering a seamless customer experience increasingly intricate.
Any breakdown of audio quality, or even reduced quality levels in communication services, can have consequences impacting not only CX, but also on brand reputation and companies' bottom line. Organizations heavily rely on various forms of communication, such as voice calls, to connect with customers, partners and employees, and the quality of these audio interactions is compromised if the underlying network is unreliable. Poor audio quality and connection rates can significantly impact speech-enabled interactive voice response (IVR) systems and voicebots.
A recent study evaluated the performance of the top three widely used automatic speech recognition (ASR) engines across a spectrum of audio quality levels. The objective was to understand the implications of poor speech quality on the accuracy of these engines. The study collected audio samples from calls made through both standard telephone and mobile networks into a contact center. The audio quality was evaluated based on industry-standard PESQ scoring algorithms, which assess audio attributes such as sharpness, background noise, audio clipping, and interference. The study comprised 12,000 recordings, with 2,000 recordings falling into each defined quality band. Speech accuracy was also measured using the word error rate (WER) metric.
The study revealed a correlation between speech quality and speech recognition accuracy, as outlined below:
- Excellent Speech Quality: 87 percent accuracy.
- Very Good Speech Quality: 83 percent accuracy.
- Good Speech Quality: 79 percent accuracy.
- Fair Speech Quality: 76 percent accuracy./li>
- Poor Speech Quality: 71 percent accuracy.
- Very Poor Speech Quality: 59 percent accuracy.
It's important to note that while WER is the industry-standard method for measuring speech recognition accuracy, it might not directly translate to the impact on customer calls. However, the data above highlights a noticeable increase in the risk of encountering issues during poor-quality calls. For instance, if a customer were to call in and say, "Can I check my balance please," and the ASR engine only recognized the words "Check" and "Balance," the call would likely be handled correctly. Conversely, if it recognized every word except "Balance," it could potentially lead to issues during the interaction. Understanding these nuances is crucial for businesses aiming to provide optimal customer experiences.
Connection Rates: A Crucial Element
Another essential aspect to consider is connection rates. In the context of the telecoms industry, connection rates refer to the speed and efficiency at which devices can establish and maintain connections to a network, and they are crucial for ensuring an effective experience during voice calls or data usage.
Unlike residential calls that give a busy signal during periods of high demand, business lines are designed to manage multiple calls simultaneously. Achieving consistent and reliable connection rates is imperative for ensuring that customers can seamlessly connect with organizations at any time. A recent study indicated that 85 percent of callers will opt against scheduling a callback and instead choose to turn to competitors.
Building on the significance of audio quality, a comprehensive analysis of telecom infrastructure performance revealed that several countries stand out as top performers in both connection rates and audio quality. Switzerland and Sweden emerge as top performers, excelling in both metrics. These mature economies prioritize network management processes, investing in network development, change management. and capacity augmentation. Surprisingly, Estonia, despite a lower connection rate, continues to exhibit strong audio quality performance, possibly due to its economic development. Italy's positive shift in audio quality suggests recent network upgrades have been successful, while Singapore's gradual improvement reflects steady progress. In contrast, China grapples with varying audio quality, potentially due to rapid service demand growth and capacity challenges.
Companies today are dedicating a significant amount of time and resources to automation and self-service, which can have a positive impact on both costs and customer experience when developed and tested correctly. Nevertheless, it is crucial to recognize that even the best speech-enabled IVRs or speech bots might struggle when the speech quality they receive is significantly compromised. Before deploying these systems, it is imperative to understand and benchmark speech quality in the market and address any issues to ensure that your systems operate effectively and ultimately deliver a superior customer experience.
Matthew Lawlor is technology vice president at Cyara.