XPENG has upgraded its auto-grade voice assistant using Microsoft custom neural voice capability, based on Neural Text-to-Speech (TTS), a feature of Azure AI. XPENG installed the new voice assistant functionality via a major over-the-air (OTA) upgrade for its P7 smart sedan customers in China. Microsoft research breakthroughs in speech, natural language and machine translation have helped significantly advance the fluency, quality, fidelity and naturalness of voice assistant technology over the past several years.

These innovations have been integrated into commercially-available speech and language capabilities within Azure Cognitive Services and other Microsoft products, so that companies like XPENG can bring richer, more engaging experiences to their customers. XPENG worked with Microsoft to overcome several key challenges to create the new cutting-edge voice assistant integration. To deal with telecommunication network jitter while the car is moving, while reducing data traffic consumption and hardware burden, and ensuring continuous high-quality speech, XPENG introduced context-specific multi-level caches, caching high-quality sound in advance and distributing it to minimize reliance on the network.

To deliver natural-sounding high-fidelity speech, XPENG uses Microsoft Azure with caching and compression to deliver XPENG's high-quality voice sampling rate of 24K Hz and quantization level of 16 bits, without overburdening the data network or the car's own CPU. XPENG also worked with Microsoft to minimize ambiguity and to optimize accuracy in voice assistant speech. As a result, the new voice assistant function has achieved new levels of lifelike voice fidelity, functionality, and scenario-specific applicability.

With these new capabilities, XPENG can deploy voice assistance in even more usage scenarios, making voice assistance an integral part of the intuitive driving experience.