This year, “AI+” has become a new quality productivity to promote economic growth and social progress, and the big model has also become the main direction for the automobile industry to embrace new quality productivity.
According to incomplete statistics, there are far more than 10 automobile brands carrying large models at the present stage, and with the strategic layout related to the artificial intelligence model of the major automobile companies, the era of “software-defined automobile” has also begun to transition to “AI-defined automobile”, the automobile industry has ushered in a new growth point, and the industrial chain has accelerated the integration.
The application of large model in automobile covers many scenarios such as vehicle networking, advanced auxiliary driving, intelligent cockpit and so on, such as creating intelligent travel system for information exchange and cooperation, perceiving environment and making judgments, creating customized services, and so on.
In many scenarios, in-vehicle voice interaction is the first stop in the field of intelligent cockpit of large models.
For automobile companies, it is very important to integrate advanced large model technology into on-board voice system, which can achieve a more intelligent experience with higher efficiency and lower cost.
In this regard, at the second Automotive artificial Intelligence Conference in 2024, Zeng Chunhua, Senior Product Director of Spitchi Intelligent Automotive Division, said that with the development of the large model of multi-mode fusion, the era of multimodal cross-domain interaction, human-like interaction, and terminal vehicle home interconnection scene fusion will eventually come, and “big model + big fusion” will redefine the human-computer interaction of intelligent vehicles.
Zeng Chunhua, Senior Product Director of Spitchi Intelligent Automotive Division, introduced pain points to big models, and Spice ingenuously created a “1mm N” model.
With the increasing demand for the ability of large models from car factories, the pain points and difficulties caused by the introduction of large models are also gradually exposed.
First of all, the introduction of large models is easy to have an impact on the existing technical links, and how to achieve the integration, scheduling and distribution of large models of different manufacturers has become a major challenge for automobile companies.
Secondly, with the increasing size of the large model, the feedback speed gradually slows down, which is particularly critical in vehicle scenarios that require millisecond response speed.
Facing the ecological integration, in order to help the car factory solve the above pain and difficulties, based on the self-developed full-link intelligent voice and language interaction technology, DFM large model and big data to build the “hub big model” framework, to create a “1x N” model, that is, the combination of “1” 100 models middle platform (central hub big model) and “N” professional big model to get on the car, and get through the front-end language processing, car use scene and back-end model ecology.
Form the ecological circle of vehicle vertical domain, make the vehicle AI experience from “voice command” to “language intelligence” + “tool intelligence”, and will absorb many top large model manufacturers to build a powerful voice interaction system, in order to gather the technical characteristics and advantages of different large models, and provide more comprehensive and accurate voice services.
Source: Spitchi, which involves upgrading the technical architecture, especially from a single large model to a multi-model converged vehicle Agent framework based on a central large model, while ensuring end-side and cloud-side distributed deployment, supporting decoupling to facilitate subsequent lightweight upgrades, and supporting plug-and-pull flexible upgrades for vertical domain models through the new large model cloud framework.
To mobilize different large models, to achieve technology landing and perfect integration with the existing voice system is an important direction of intelligent cockpit development.
The process of model fusion mainly involves two directions: first, the vertical fusion of large and small models.
The large model has powerful capabilities such as semantic understanding and task reasoning, but it can not completely replace the small model.
In some basic and vertical industries, small models have the characteristics of accurate classification and fast response, which are still indispensable.
Therefore, it is necessary to realize the cooperation between the small model and the large model to ensure the proper scheduling relationship between them.
Secondly, it is the horizontal ecological fusion of large models, such as focusing on the fusion of large models in different fields, such as multi-mode understanding, text generation, picture generation, audio generation and so on.
This means that among the many large models, choose the most suitable for a new and more intelligent voice interactive understanding around the car scene, including the scene recommendation of the large model, and the personification and emotional fusion.
Focusing on multiple vehicle scenarios and reshaping the voice interactive experience, the penetration rate of voice control functions of domestic passenger cars continues to rise.
According to Gaishi Automotive Research Institute, the cumulative penetration rate of voice control functions exceeded 75% in 2023.
It is estimated that by 2025, the penetration rate of domestic voice interaction products will reach 92%, and the market size will reach 2.
3 billion yuan.
With the improvement of users’ demand for in-vehicle voice interactive experience, the carrying capacity of each voice function is showing an obvious upward trend.
Spitchi is actively exploring on-board application scenarios, paying special attention to the high-frequency needs of users.
Users mainly pay attention to navigation, car control, multimedia, telephone and weather inquiry functions in the car.
Among them, the demand for custom wake-up words is particularly prominent, which reflects users’ expectations for personalized and intelligent voice interaction.
, image source: Sibichi, in view of the above high-frequency scenes, Spichi focuses on skill reconstruction.
Travel is an important part of people’s daily life, whether it is commuting, business trip or leisure and entertainment, itinerary planning is a key link.
In this process, navigation has become an important application scene of vehicle voice interaction.
The traditional navigation process is tedious, and the introduction of the Spitz model can achieve a convenient experience of “one step in place”.
Through the travel planning function, users can quickly generate a travel plan and navigate to the destination with a simple sentence.
Image source: Spichi, car voice interaction also involves many other scenes, such as car control, entertainment and so on.
Especially when there is a vehicle failure or traffic accident, users need to get the solution quickly.
Spitchi large model can combine rich data and knowledge to provide users with timely and accurate feedback and suggestions.
In addition, the emotional demand of users for vehicles is also growing gradually.
Spitchi model can provide more intimate and intelligent services combined with the personalized needs of users.
According to reports, at present, version 7.
0 of Tianqin Voice Assistant, which is based on the large model scheme of multi-mode fusion, can realize multiple original landing applications, such as free switching of multi-user settings, large model itinerary planning, cross-domain context intention understanding, one sentence variable wallpaper and so on.
The global end-to-end speed and multi-round conversation speech synthesis speed can be controlled within 3 seconds, and the overall accuracy of rejection + response is 92.
7%, of which the accuracy of non-chat is more than 98%, which greatly reduces the impact of user chat sound on cockpit interaction.
Adhere to the accumulation of conversational AI technology, Spitz continues to drive the car voice interaction scene, Spichi is a domestic professional dialogic artificial intelligence platform enterprise, based on self-developedFull-link intelligent dialogue system customized development platform, language computing model DFM, artificial intelligence voice chip, around the “cloud + core” layout, to provide artificial intelligence technology and product services with the combination of software and hardware for the Internet of things, digital government and enterprises.
The core advantages of Spitzer are the full-link voice interaction technology, the construction ability of soft-hard integrated man-machine dialogue system and the customization ability of large-scale dialogue system.
With the accumulation of core technology in the industry, Spitchi combines the full-link intelligent dialogue customization platform (DUI platform) with the DFM-2 model to support customers to build personalized voice interaction solutions with “thousands of people and thousands of faces” independently, so as to realize large-scale, high-quality and personalized artificial intelligence system customization.
In recent years, with the goal of improving the accuracy of speech recognition, the naturalness of interaction and the intelligent level of the system, Spichi has achieved a number of innovations and breakthroughs: it has led and participated in the development of 70 national / industry / group standards, covering speech synthesis, speech recognition, voice print recognition, machine learning, intelligent ability levels, natural language processing, etc.
It has nearly 100 global original technologies, nearly 1500 pieces of intellectual property rights, 23 excellent certification related to products and technologies, and has won the championship in international evaluation many times.
Has been deeply involved in the automotive industry for many years, and Spice has many years of experience in on-board voice.
From the initial vehicle voice system to the current multi-mode fusion large model scheme, the core is always to meet the needs of users.
In 2022, Spitchi Automotive Voice interaction Program (Tianqin Assistant) passed the trusted AI evaluation of China Institute of Information and Communication, and obtained L9 intelligent certification, which is the highest intelligent grade certification of in-vehicle voice interaction products known in China.
In 2023, the application of Spitchi large model in intelligent cockpit was successfully selected into the “White Paper on the Application of large Model Industry” jointly issued by China quality Certification Center and Zhongguancun Zhiyong Institute of artificial Intelligence, highlighting the landing ability of Spitchi large model in practical application scenarios.
The first domestic standard for large-scale automobile model compiled by Spitchi has also been officially released recently (“Industry-oriented large-scale pre-training model technology and application evaluation method part 4: automobile” standard), focusing on the high-quality development of the automotive industry.
Promote the upgrading and optimization of large-scale automobile model products.
In 2024, Spitchi upgraded the full-scene application of large model technology and launched a sea strategy.
In the automotive field, on the one hand, it increased its localization cooperation with overseas car companies, on the other hand, it accompanied Chinese new energy car companies to “go out”.
Led by Spitchi, the international standard of vehicle-mounted multi-tone area voice interaction (the framework and requirements of Framework and requirement for in-vehicle multi-region intelligent speech interaction system/ vehicle-mounted multi-tone area voice interaction system) proposed by China Institute of Information and Communications and China Telecom was discussed and established by the 16th Research Group (SG16) of the Bureau of Standardization of the International Telecommunication Union (ITU-T) in France in April.
This is also the first new generation of automobile voice interaction international standard led by a Chinese company in this field.
At present, Spitchi has cooperated with nearly 60 automobile brands, including new power brands, independent brands and international giants, and its products have been applied to more than 160 mass-produced models, with a cumulative installed capacity of more than 10 million vehicles.
In the field of on-board voice, Spice has a remarkable growth rate, and the domestic new energy vehicle-borne voice has the first market share, and it is among the Top 3 in the whole vehicle voice industry.
Tu Yuan: Spitchi, Zeng Chunhua said that Spitchi is committed to working with large model manufacturers and car manufacturers to jointly promote the development of in-vehicle voice interaction technology.
At the same time, Spitchi will also pay attention to the cutting-edge technologies such as multimodal interaction and ecological integration, and constantly improve the intelligent level of in-vehicle voice interaction.
Spitchi hopes to create a vehicle vertical ecosystem through the scheduling and fusion capabilities of the central model with the help of full-link voice technology, so as to bring users a more convenient and intelligent travel experience! , return to the first electric network home page >.
