Fraunhofer’s upHear Voice Quality Enhancement reduces background noise and acoustical echoes, ensuring that Yandex’s new smart speaker, Yandex.Station and the embedded intelligent assistant Alice, can hear intelligible verbal commands, even when the speaker is streaming music at the same time.
“Fraunhofer provided a software solution that applies advanced digital signal processing techniques to enhance the signals collected by the microphone array technology for far-field voice capturing, Yandex.IO. This ensures that Yandex’s highly reliable keyword spotter and speech recognizer receive a clean voice signal,” says Manfred Lutzky, Head of Audio for Communications at Fraunhofer IIS. “Our upHear technology removes interfering sounds captured by the smart speaker’s microphones, extracts the user’s voice and cancels out acoustical echoes that would otherwise make it impossible for the speaker to understand the user’s request.”
Fraunhofer’s upHear Voice Quality Enhancement solution supports full-bandwidth audio and is a suitable voice enhancement step for a wide range of applications, including natural language understanding in mobile and smart home devices. It is a fully integrated and flexible solution which enables full-duplex conversations and barge-in support with outstanding audio quality. This is achieved by combining advanced multichannel source localization and beamforming techniques with echo- and noise-reduction algorithms.
“We tested multiple voice enhancement solutions working together with our far-field microphone array technology Yandex.IO and found the Fraunhofer upHear technology to be the most effective in bringing Alice from the smartphone to the smart home,” explains Konstantin Kruglov, Director of Experimental Products at Yandex. “Whether you are standing right in front of the device, or farther away in the room, Alice will understand you, even under challenging acoustic conditions.”
Yandex’s AI assistant Alice listens for and understands natural Russian-language requests and provides contextually relevant answers using Yandex’s search engine and services that include weather, news, maps, transport and music streaming services. In developing Alice, Yandex built on its knowledge of the more than 50 million monthly users who interact with its services, plus its machine-learning capabilities and 20 years of experience with Russian users and the Russian language.
Using Yandex’s proprietary speech recognition technology, Alice delivers world-best accuracy for spoken Russian-language recognition and understands speech with near human-level accuracy. Alice also synthesizes voice responses that sound natural and authentic. With the new integration into Yandex.Station and several other updates, Alice’s ecosystem is expanding to offer Russian users an even more advanced experience.
www.iis.fraunhofer.de
yandex.com
