賽輪思AI與聯(lián)發(fā)科合作推出基于英偉達技術(shù)的端側(cè)多模態(tài)語言模型

在端側(cè)配備智能護欄功能的多模態(tài)模型可實現(xiàn)更無縫、更安全的車內(nèi)交互，解鎖強大功能，遠超單純的信息娛樂體驗

上海及馬薩諸塞州伯靈頓，2025年4月23日——Cerence Inc.（NASDAQ: CRNC）（“賽輪思AI”），全球?qū)υ捠紸I用戶體驗創(chuàng)新的領(lǐng)導(dǎo)者，今日宣布與聯(lián)發(fā)科（MediaTek）達成全新戰(zhàn)略合作關(guān)系，并擴大與英偉達（NVIDIA）的現(xiàn)有合作，共同推出下一代CaLLM™ Edge，一款專為汽車用戶交互體驗設(shè)計的嵌入式小語言模型（Small Language Model，SLM）。憑借由NVIDIA AI Enterprise軟件驅(qū)動的全新多模態(tài)功能，CaLLM Edge將使車內(nèi)交互比以往任何時候都更智能、更具感知能力、更人性化、更安全。這款多模態(tài)端側(cè)小語言模型（SLM）已無縫集成到Cerence xUI™——全新一代端云混合、智能體化的AI平臺，并將于2025上海車展上首次亮相。

此次發(fā)布的信息源于聯(lián)發(fā)科與英偉達的持續(xù)合作，聯(lián)發(fā)科的汽車級天璣汽車座艙芯片組集成了英偉達的GPU、人工智能和圖形技術(shù)。如今，這兩家公司已與賽輪思AI攜手推出多模態(tài)端側(cè)解決方案，為汽車制造商提供卓越性能并具有成本效益，同時為駕駛者和乘客解鎖強大的新功能。利用聯(lián)發(fā)科的汽車座艙解決方案，并率先在NVIDIA DRIVE AGX上進行優(yōu)化，CaLLM Edge的先進多模態(tài)功能通過雙處理平臺，針對車內(nèi)部署進行高效率微調(diào)，實現(xiàn)了車內(nèi)車外的全新情境感知。

聯(lián)發(fā)科展臺上展示由賽輪思AI 提供技術(shù)的車載AI語音助理

賽輪思AI產(chǎn)品與技術(shù)執(zhí)行副總裁Nils Schanz（左四）和聯(lián)發(fā)科企業(yè)副總裁兼汽車業(yè)務(wù)總經(jīng)理Mike Chang（左五）在上海車展現(xiàn)場

上海車展標(biāo)志著Cerence xUI首次在NVIDIA DRIVE AGX上運行，并集成到模擬量產(chǎn)的汽車設(shè)計中，該設(shè)計復(fù)制了實車的多座環(huán)境，為駕駛者和乘客帶來無縫、個性化的交互體驗。關(guān)鍵創(chuàng)新包括：

車內(nèi)智能提升——想象一下，您的汽車能感知到您何時在與它交談，何時沒有。您可以要求推薦意大利餐廳，然后問乘客：“第二家怎么樣？”車載助理會知道要等待，再繼續(xù)對話。此外，它還能同時處理駕駛者和乘客的指令，例如，“打開我的車窗”，隨后說“我也要”?；蛘撸胂笠幌?，您的汽車可以在您短途旅行時照看您的寵物，或者在您駕駛時回答有關(guān)后座小孩的情況。

車外情境感知增強——想象一下，一個助理不僅能提供車內(nèi)信息，還能提供車外世界的信息。例如，它能解釋或翻譯路標(biāo)，識別路邊建筑，甚至提供有關(guān)廣告牌上有趣事物的詳細信息。Cerence xUI可以結(jié)合汽車傳感器數(shù)據(jù)以及內(nèi)部和外部攝像頭視頻流，利用CaLLM Edge多模態(tài)SLM，在端側(cè)實現(xiàn)實時輸入處理——所有這些應(yīng)用場景都將首次在上海車展上展示。

端側(cè)智能護欄——賽輪思AI已針對NVIDIA DRIVE AGX平臺開發(fā)了優(yōu)化的汽車智能護欄技術(shù)，該技術(shù)整合了NVIDIA NeMo Guardrails功能，確保其驅(qū)動的系統(tǒng)在連接條件受限時，仍能安全可靠地提供品牌專屬的車內(nèi)交互體驗。這些智能防護機制有助于最大程度減少分心、防止誤用，并確保語音交互始終保持情境感知且符合安全規(guī)范。

除了這些為駕駛者帶來的新功能外，CaLLM Edge還為汽車制造商提供了關(guān)鍵優(yōu)勢。多模態(tài)處理的主要工作在端側(cè)完成，這意味著所需的云端帶寬更少，汽車制造商成本更低。此外，使駕駛者能夠訪問車外信息，并提供先進的生成式人工智能功能，為車企解鎖了新的訂閱服務(wù)和收入來源。

賽輪思AI產(chǎn)品與技術(shù)執(zhí)行副總裁 Nils Schanz 表示：“隨著我們不斷突破CaLLM系列云端和嵌入式語言模型的極限，我們繼續(xù)提升滿足各種汽車制造商及其特定信息娛樂平臺需求的能力。通過與聯(lián)發(fā)科和英偉達合作，我們?yōu)镃aLLM Edge注入了實時感知和情境感知能力，重新定義了車內(nèi)體驗。”

聯(lián)發(fā)科企業(yè)副總裁兼汽車業(yè)務(wù)總經(jīng)理Mike Chang表示：“賽輪思AI的CaLLM Edge與聯(lián)發(fā)科天璣汽車平臺的結(jié)合，為車輛中強大的多模態(tài)、生成式人工智能驅(qū)動體驗奠定了堅實基礎(chǔ)。聯(lián)發(fā)科天璣汽車平臺在端側(cè)支持Cerence LLM和SLM，而我們新推出的配備雙AI引擎的C-X1非常適合可靠地同時運行多個AI任務(wù)。這一合作是我們汽車生態(tài)系統(tǒng)的重要補充，將使全球汽車品牌能夠自信地加速下一代設(shè)計。”

英偉達汽車業(yè)務(wù)副總裁Rishi Dhall表示: “賽輪思AI與聯(lián)發(fā)科的合作基于公司部署NVIDIA人工智能和加速計算以提升整體車內(nèi)體驗。雙方合作使汽車制造商能夠更好地控制語音體驗的定制，而端側(cè)NVIDIA NeMo護欄的集成在安全、可擴展的車內(nèi)人工智能可成為新的行業(yè)標(biāo)桿。”

賽輪思AI于4月23日至5月2日在上海車展上與聯(lián)發(fā)科和英偉達共同展示其產(chǎn)品。感興趣的參觀者可以在8.2號館8BD002號展臺體驗Cerence xUI 實車演示。該解決方案將在聯(lián)發(fā)科展臺——8.2號館8BE006號展臺上展示。

如需了解更多關(guān)于賽輪思AI的信息，請訪問www.cerence.ai 或www.cerence.cn 。

關(guān)于Cerence Inc.

賽輪思（Cerence Inc，NASDAQ: CRNC）是全球汽車行業(yè)及移動出行領(lǐng)域在創(chuàng)造直觀、無縫且基于人工智能的用戶體驗方面的領(lǐng)導(dǎo)者。憑借超過二十五年在語音、生成式AI與大語言模型領(lǐng)域的創(chuàng)新成果和技術(shù)優(yōu)勢，賽輪思為駕駛者與乘客精心打造了集成的車載體驗，使每一次出行都更加安全、互聯(lián)和愉悅。至今，已有逾 5億輛汽車采用了賽輪思技術(shù)。賽輪思與全球頂尖的汽車制造商、移動出行OEM及科技領(lǐng)軍企業(yè)緊密合作，共同推動下一代用戶體驗的發(fā)展。賽輪思中國總部設(shè)于上海，公司業(yè)務(wù)遍布全球，其全球團隊致力于持續(xù)突破人工智能的創(chuàng)新邊界。如欲了解更多信息，請訪問www.cerence.ai  或 www.cerence.cn 。

此新聞稿為翻譯本，所有內(nèi)容以Cerence英文版新聞稿為準(zhǔn)。

Cerence AI and MediaTek Partner to Introduce Multi-Modal Language Models Running on the Edge, Built on NVIDIA

Multi-modal models with smart guardrailing at the edge enable more seamless and safer in-car interactions, unlocking powerful capabilities that go far beyond pure infotainment

SHANGHAI and BURLINGTON, Mass., April 23, 2025 – Just ahead of Auto Shanghai 2025, Cerence Inc. (NASDAQ: CRNC) (“Cerence AI”), a global leader pioneering conversational AI-powered user experiences, today announced a new partnership with MediaTek and expanded collaboration with NVIDIA to deliver the next generation of CaLLM™ Edge, its embedded small language model (SLM) purpose-built for automotive user experiences. With new multi-modal capabilities powered by NVIDIA AI Enterprise software, CaLLM Edge will now make in-car interactions smarter, more perceptive, more human, far safer and more secure than ever. This multi-modal edge SLM is seamlessly integrated into Cerence xUI™, the company’s hybrid cloud/embedded agentic AI platform, and will be showcased for the first time at Auto Shanghai 2025.

Today’s announcement stems from an ongoing collaboration between MediaTek and NVIDIA, with MediaTek’s automotive-grade Dimensity Auto Cockpit chipsets integrating NVIDIA’s GPU, AI, and graphics technologies. Now, the companies have teamed with Cerence AI on a multi-modal edge solution that delivers top performance and cost efficiency for automakers and unlocks powerful new features for drivers and passengers. Leveraging MediaTek’s automotive cockpit solution and optimized first on NVIDIA DRIVE AGX, CaLLM Edge’s advanced multi-modal capabilities are fine-tuned for in-vehicle deployment and high efficiency via a dual processing platform, delivering a new level of contextual awareness – both inside and outside the vehicle.

Auto Shanghai marks the first public demonstration of Cerence xUI running on NVIDIA DRIVE AGX, integrated into a production-ready automotive reference design that replicates a real car’s multi-seat environment, enabling seamless, personalized interaction for drivers and passengers alike. Key innovations include:

Increased intelligence inside the car – Imagine your car knows when you’re talking to it – and when you’re not. You could ask for recommendations for Italian restaurants, then ask your passenger: “What about the second one?” – and the assistant knows to wait to jump back into the conversation. Plus, the assistant can take direction from driver and passengers at the same time – for example, “Open my window,” followed by “Same for me.” Or, imagine your car can keep an eye on your pet while you run a short errand, or, while you’re driving, answer questions about how your toddler is doing in the back seat.

Additional context from outside the car – Imagine an assistant that provides information about not only what is inside the vehicle, but also in the world outside. For example, it can explain or translate road signs, identify roadside buildings, and even offer details about something interesting spotted on a billboard. Cerence xUI can leverage car sensor data and both interior and exterior camera feeds in combination with the CaLLM Edge multi-modal SLM, enabling streaming input on the edge – all of which will be shown for the first time at Auto Shanghai.

Smart guardrailing on the edge – Cerence AI has developed automotive guardrails optimized for NVIDIA DRIVE AGX, helping ensure Cerence-powered systems can safely and reliably handle the nuances of brand-specific in-car interaction, even in conditions with limited connectivity. As part of its comprehensive hybrid architecture, Cerence AI will be delivering this embedded guardrail with NVIDIA NeMo Guardrails, both to complement its cloud variant as well as help minimize distractions, prevent misuse, and ensure voice interactions stay context-aware and compliant with safety norms.

Beyond these new capabilities for drivers, CaLLM Edge also offers key benefits for automakers. The heavy lifting of the multi-modal processing happens at the edge, meaning less cloud bandwidth required and reduced costs for OEMs. In addition, enabling drivers to access information outside the car and delivering advanced, generative AI-powered features unlocks new subscription-based services and revenue streams for OEMs.

“As we continue to push the limits of our CaLLM family of cloud and embedded language models, we continue to improve our ability to meet the needs of a wide variety of automakers and their specific infotainment platforms,” said Nils Schanz, EVP, Product & Technology, Cerence AI. “By teaming with MediaTek and NVIDIA, we’ve supercharged CaLLM Edge with real-time perception and contextual awareness that redefines the in-vehicle experience.”

“The combination of Cerence’s CaLLM Edge with MediaTek Dimensity Auto platforms creates an incredible foundation for powerful multi-modal, generative AI-driven experiences in vehicles,” said Mike Chang, Corporate Vice President and General Manager of Automotive Business of MediaTek. “MediaTek Dimensity Auto supports both Cerence’s LLM and SLM on the edge, and our new C-X1 featuring dual AI engines is ideal for the task of reliably running multiple AI tasks simultaneously. This partnership is a key addition to our automotive ecosystem and will enable global automotive brands to accelerate their next-generation designs in confidence.”

“Cerence’s collaboration with MediaTek builds on the company’s work to deploy NVIDIA AI and accelerated compute to improve the overall in-vehicle experience,” said Rishi Dhall, Vice President of Automotive at NVIDIA. “The work now gives automakers greater control to customize voice experiences, and our work to integrate NVIDIA NeMo Guardrails at the edge has the power to set a new industry benchmark in the evolution of safe, scalable in-car AI.”

Cerence AI will demonstrate its offering with MediaTek and NVIDIA at Auto Shanghai, taking place April 23-May 2. Interested attendees can experience Cerence xUI in action in Hall 8.2, booth number 8BD002. The solution will also be shown at MediaTek’s booth, also in Hall 8.2, at booth number 8BE006.

To learn more about Cerence AI, visit www.cerence.ai, and follow the company on LinkedIn.

About Cerence Inc.

Cerence Inc. (NASDAQ: CRNC) is a global industry leader in creating intuitive, seamless, AI-powered experiences across automotive and transportation. Leveraging decades of innovation and expertise in voice, generative AI, and large language models, Cerence powers integrated experiences that create safer, more connected, and more enjoyable journeys for drivers and passengers alike. With more than 500 million cars shipped with Cerence technology, the company partners with leading automakers, transportation OEMs, and technology companies to advance the next generation of user experiences. Cerence is headquartered in Burlington, Massachusetts, with operations globally and a worldwide team dedicated to pushing the boundaries of AI innovation. For more information, visit www.cerence.ai.