1. Foreword: iFLYTEK Spark Large Model V3.5 Spring Update strengthens long text processing and multimodal interaction capabilities
In this era of rapid information volume, the ability to quickly and efficiently obtain and process information has become a rigid need for professionals, students, and researchers.
In order to meet this urgent need, iFLYTEK has launched the industry's first large model of long text, long graphics and long voice, helping users easily grasp the initiative of information in the era of information flooding, and the problem of information overload has been fundamentally solved.
This is the new version of iFLYTEK Xinghuo cognitive model V3.5, which focuses on improving the processing capabilities of long text, long graphics and text, and long speech, and adds functions such as multi-emotional super-anthropomorphic synthesis and one-sentence voice replication.
In terms of processing long texts, iFLYTEK Xinghuo Large Model V3.5 can quickly integrate a large amount of information from multiple channels, provide more accurate answers for different industries, and significantly improve the speed of file uploading, knowledge Q&A and text generation efficiency.
For non-traditional text acquisition methods, such as screenshots, PPT pictures of academic conferences or recordings of important conversations, iFLYTEK Xinghuo can handle complex layouts and recognize special professional symbols through its image and text recognition and audio and video recognition capabilities to meet the needs of specific fields.
In addition, users can also experience similar functions to dialogue with real people through the iFLYTEK Xinghuo APP, including happiness, apologies, comfort, coquettishness, confusion and other emotions.
The new one-sentence sound forking feature allows users to create personalized sound content.
So, what exactly is the experience of these new features?
2. Long text: full-dimensional upgrade of information insight ability An all-round helper in the professional field
In terms of long text recognition and creation, iFLYTEK Xinghuo V3.5 is new in spring, mainly updating the ability of long text Q&A, long text summary and long text generation, the ability to accurately locate and answer specific questions in massive information, the ability to extract core ideas and the logic of text generation have been greatly enhanced.
1. Long text Q&A
(1) Daily contracts
I believe that many people will encounter some thorny problems when writing contracts, which more or less hinder our work efficiency and the legal validity of contracts.
In this regard, iFLYTEK Xinghuo has specially developed the "Xinghuo Contract Assistant", a small component, which provides a comprehensive solution to problems such as difficulty in contract writing and understanding.
For example: a lease contract
Question 1: Help me draft a rental agreement
Using the Spark Contract Assistant, you can directly let it help us draft the contract, in the contract, contains detailed information about the two parties, deadlines and schedules, payment terms, ways to resolve disputes, liability for breach of contract, additional terms, etc., with the help of this template, we only need to make simple adjustments and fills in according to the specific situation of the actual cooperation, which is very convenient.
Secondly, we can send the prepared contract to the Spark contract assistant in the form of a PDF file for review, and ask whether there are unreasonable provisions and omissions in the contract to ensure the rigor and completeness of the contract.
Of course, if you have any questions about the content of the contract, you can also ask the "Spark Contract Assistant" directly.
We can directly upload the documents that need to be identified through the "Documents" button, and ask the questions that need to be asked below.
For example, we asked them the following two questions:
Question 2: What are the terms in the contract that are unfavorable to Party B?
Question 3: What problems are encountered and the contract will be terminated.
It can be seen that the Spark Contract Assistant will deeply analyze the terms of the contract and provide comprehensive answers to our questions, which will be presented in a clear and clear digital form, helping us accurately grasp the specific details of the contract to avoid any misunderstandings and potential risks.
Of course, it is inevitable that we will encounter borrowing matters in our lives, whether it is a temporary turnover between individuals or a fund allocation between enterprises, the best solution is to draft a clear and legally effective loan contract, and we must ensure that the matters in the contract are comprehensive and specific.
So, the Spark Contract Assistant can also help you.
For example: a loan contract
Question 1: Please outline the types of loans offered in this contract and their respective purposes.
Question 2: List the repayment methods provided in this contract and explain how Party A should repay the loan according to the contract.
Question 3: Give me the issues in the contract that I need to pay special attention to.
It can be seen that the "Spark Contract Assistant" widget can be used to automate the contract writing and review process, ensure the legal accuracy and compliance of the contract content, and greatly improve work efficiency.
Moreover, the biggest advantage of this widget is that you can ask any professional questions in the contract directly, and it can simplify the professional questions into easy-to-understand language or explanations, and non-professionals can understand and grasp the key terms of the contract.
(2) Industry norms
Regulations and national industry norms play an important role in our lives, and if we have difficulty reading or understanding them, then these regulations become "scrap letters" for us.
However, with iFLYTEK Xinghuo, it's different.
It can identify these professional document files, dig deep and understand the content in these documents, deeply understand their meaning, and transform them into concise and clear text content, which is convenient for us laymen to read and understand.
In this regard, we also used several professional specification documents in the industry to test iFLYTEK Xinghuo's ability to understand words.
(1) Insurance
(2) Highways
(3) Banks
It can be seen that iFLYTEK Xinghuo is not only able to accurately identify and understand these documents and extract the main points of the problem, but also to convert them into easy-to-understand words and express them in a concise and clear way, so that those of us who are not experts can understand it at a glance.
(3) Reading and entertainment
In terms of entertainment, iFLYTEK Xinghuo has also shown its strong comprehension and reading ability, and for book lovers, iFLYTEK Xinghuo can help them better understand and enjoy the fun brought by novels.
Specifically, iFLYTEK Xinghuo can conduct in-depth analysis and understanding of the text in the novel through natural language processing technology and machine learning algorithms. The plot, characters, and dialogues in the novel are all able to accurately identify and understand the meaning and emotion in them.
We also selected the classic novels of Yu Hua and Jin Yong to test them.
(1) "Alive" - Yu Hua
(2) "Snow Mountain Flying Fox" - Jin Yong
We can see that iFLYTEK Xinghuo answered the questions we raised accurately.
2. Long text summaries
Text summarization has always been iFLYTEK's strong point, and this time in the V3.5 iteration, the ability to summarize long text has been added, for example, in scientific research scenarios, by identifying and understanding a doctoral dissertation of nearly 150,000 words, and asking questions to it.
Research Scenario: Research on the Impact of Digital Economy on Labor and Employment
Leveraging this ability, iFLYTEK Xinghuo is able to extract key concepts, themes, and arguments from large amounts of information, providing users with a concise overview without losing the main meaning of the original text.
3. Long text generation
3. Long graphics and texts: Accurately analyze the content of graphics and texts and easily identify key elements
In the face of pictures or long texts containing rich visual information, iFLYTEK Xinghuo's image and text recognition model shows an extraordinary ability, which can not only accurately analyze complex layout layouts, but also be widely applicable to various environments and situations, and recognize a variety of key elements and cross-domain symbols.
We can directly upload the pictures that need to be identified through the "Graphic and Text" button, and ask the questions that need to be asked below.
For example: Everyday life scene Identification manual
(1) What environment does this radiator need to be used in, and what environment does it need to be stored?
It can be seen that through text recognition, iFLYTEK Xinghuo has fully understood the meaning expressed in the manual, and through our questions, it can give accurate answers to questions.
(2) How does this radiator switch the light language?
In the face of this kind of picture + text manual, it is more difficult to test the processing ability of the large model for image and text recognition.
Through the test, it can accurately identify the text content with picture partitions, and answer the questions accurately in the face of the questions raised.
It can even recognize the text in the image with a very impressive degree of accuracy.
Fourth, long voice: break through the boundaries of audio and video, accurate identification, and summarize in place
In this spring update, the iFLYTEK Xinghuo model also added the ability to recognize audio and video, and we can freely upload audio and video files, mainly for learning and work scenarios such as meeting recordings, conversations, learning, interview videos, lecture videos, etc.
In the interface, we select the "Audio and Video" button to directly upload the audio or video file, and the large model can be deeply recognized.
We have also prepared an audio file and a video file to test its ability to recognize audio and video and answer questions.
(1) Recording: Luo Xiang talks to Yu Hua
Question 1: Why is Lu Xun a book written by a patient?
Question 2: In the book "Brothers", how does the author reveal the ills of society through his own experiences and observations?
In the face of the recognition of audio files, iFLYTEK Xinghuo will transcribe the audio content, and we ask it to answer the audio content by asking questions, and the final answer does not deviate from the topic, nor is it ambiguous, but gives the answer to the question directly, from here we can clearly see the excellent performance of iFLYTEK Xinghuo in speech recognition and language understanding.
(2) Video: Fan Zhendong WTT Incheon Championship interview video
We also selected a video of Fan Zhendong's WTT Incheon Championship interview for iFLYTEK Xinghuo to identify and give it a brief overview of the content of the video.
Question: A brief overview of what the entire video is about
In about 30 seconds, iFLYTEK Xinghuo can accurately identify a video of about 5 minutes, understand its connotation, and summarize it into a brief overview of about 200 words.
This technology has a wide range of applications and can be used in a variety of scenarios, such as news reports, educational lectures, and business speeches.
For example, journalists can use iFLYTEK Spark to quickly generate news summaries, saving a lot of time and effort, teachers can use it to help students understand and remember class content, and business people can use it to quickly access key information from meetings or presentations.
What's more, the documents, videos, pictures, audio, and other content we have uploaded will be temporarily stored in the "personal space", which is convenient for quick search and knowledge quizzes in the future.
Personal Space can be found in the upper right corner of the interface and in the dialog box at the bottom.
5. iFLYTEK Xinghuo voice model: 3 kinds of timbres You can inject your own voice into it
1. Multi-emotional superanthropomorphic synthesis
The Xunfei Xinghuo V3.5 updated in February already has the ability of voice interaction, and in the voice interaction, two timbre outputs are injected, namely Ling Xiaoyue (female voice) and Ling Feiyi (male voice).
Two months later, the new version added Tianjin dialect (Zijin), forming three kinds of multi-emotional super-anthropomorphic synthesized sounds, injecting soul into voice interaction.
2. A one-sentence sound reproduction
In this update, you can also record your own voice into it, and you only need "one sentence" to reproduce your own voiceprint.
Before recording the voiceprint, we need to select the gender, and then we can move on to the recording.
The recording process is very concise, and you only need to read the above paragraph.
Once the voiceprint is recorded, you can find your voice in the "Voice Talents I Created" and customize the name.
After choosing, we will be able to hear our own voice when we use voice interaction in the future!
6. Summary: A new milestone in intelligent information processing
The iFLYTEK Spark V3.5 spring update mainly focuses on long text Q&A, summary generation, and audio and video recognition, which are significantly enhanced, aiming to improve the efficiency of users in processing complex information scenarios.
This new feature is ideal for understanding and organizing legal contracts, industry regulatory documents, and scientific research materials.
The newly introduced "Spark Contract Assistant" widget fully assists users in drafting and reviewing contracts quickly, while providing easy answers to professional questions to ensure legal accuracy and compliance.
In addition, the long text summarization function has been optimized to effectively extract core ideas from large texts, simplifying scientific research and academic work. Long text generation enhances the logic and coherence of the text, supporting a wide range of content creation needs.
Advances in image and text recognition and audio and video recognition allow users to deeply analyze multimedia content and provide rich scene applications.
In addition, the multi-emotional voice interaction experience and voice replication function greatly improve the quality of interaction between users and the system, make the user's interaction more natural and personalized, and make the voice interaction function no longer a decoration.
On the whole, the new version of iFLYTEK Xinghuo significantly improves the efficiency of knowledge acquisition and learning, and effectively meets the rigid needs of different users in various scenarios.
With the help of iFLYTEK's Xinghuo cognitive model, even in the era of information flooding, we can easily filter and organize the necessary information, without worrying about too much information and difficult to deal with.
The launch of this new version of the large model is particularly important for professionals, students, and researchers, and can also provide us with strong support in an information-intensive environment at critical times.