Call Center Management Featured Article
Google Steps Up Contact Center AI Game
Google (News - Alert) has finally made significant strides with its long anticipated Contact Center AI, announced a year ago at the company’s Cloud Next conference. Designed to aid customer service agents and promote better customer interactions, the toolkit now contains a host of new features designed to improve speech recognition accuracy.
Powered by machine learning, Google’s Contact Center AI uses the Dialogflow conversational experiences development platform in tandem with its Cloud Speech-to-Text suite to bolster customer interactions on the phone. The company has added a new Auto Speech Adaptation feature, currently in beta, that focuses on instances in which the speech recognition system might confuse similar-sounding words. The new feature uses context, such as phrases used in training or agent-specific information, to improve responses in specific situation. The process is called speech adaptation, and is one of the many machine learning tools the Contact Center AI is taking advantage of.
“We’re excited to see how these improvements to speech recognition improve the customer experience for contact centers of all shapes and sizes,” wrote Dan Aharon and Shantanu Misra, product managers for Contact Center AI, in a blog post. “We’re constantly adding more quality improvements to the roadmap — an automatic benefit to any IVR or phone-based virtual agent, without any code changes needed — and will share more about these updates in [the] future.”
The move comes as cloud competitor Microsoft (News - Alert) recently announced a controversial $1 billion AI investment. And a new report from Orbis Research names IBM (News - Alert), Google, AWS and Microsoft among the key players fueling the global call center AI market, boosting competition and innovation in the call center AI space.
In other call center enhancements to its AI model, Google recently launched premium speech-to-text models for specific use cases. A phone model that is optimized for two- to four-person conversations was made generally available in February, with Google claiming it had 62-percent fewer transcription errors than its previous version. That model has been optimized even more to handle short utterances in U.S. English, and is now 15-percent more accurate.
Google has also debuted a number of new AI features in beta, including “richer” manual speed adaptation and entity classes, expanded phrase limits and endless streaming. Cloud Speech-to-Text had previously supported streaming audio in one-minute increments, but can now process sessions up to five minutes and resume streaming where previous sessions leave off. The solution also now supports MP3 files.
SpeechContext, the collection of the AI’s speech-to-text settings, now includes prebuilt classes designed to handle concepts like digit sequences, addresses, number and money denominations. The settings also optimize ASR to immediately create a list of words, while also helping adjust speech adaptation strength to cut down on the number of false positives. SpeechContext has also been scaled to support up to 5,000 phrase hints per API, which increases the probability that uncommon words and phrases will be captured by ASR.
Edited by Maurice Nagle