There has always been a conflict between audio and video experience and cost. How to maximize user experience and achieve the best balance between the two under limited resource conditions is a long-term challenge for companies that provide on-demand solutions. At the Beijing station in 2023, LiveVideoStack invited Ma Qian, the person in charge of the VOD basic technology of Volcano Engine, to share. From the dialectical perspective of data science, she explained that on-demand optimization is a non-zero-sum game relationship between "cost and experience". In the following months, she continued to explore the technical direction of image quality optimization and interactive videos, and made deeper breakthroughs in the relationship between cost optimization and experience impact.
The following is a conversation we had with her recently -
Interviewees |
Planned and edited by Ma Qian | Xin Wang
LiveVideoStack: Hello Maqian, I am very glad to invite you again at LiveVideoStackCon 2023 Shanghai Station. Before the official interview, can you share which technologies you have been paying attention to recently?
Ma Qian: Recently, I will pay attention to some image quality optimization and interactive video technologies. With the gradual development of multimedia technology, the dimension of video information expression is gradually expanding, such as color gamut width, definition, spatial dimension and so on.
Picture quality and gameplay are some of the more popular directions in recent years. We often talk about "picture quality is faith". obvious. How to better understand image quality and balance the relationship between performance, cost, and image quality is a direction that I want to further explore recently.
The other is interactive videos. In the future, the game play scheme will definitely make breakthroughs in content genres, and immersion and interactivity are also mainstream trends. At present, there are no explosive products in the industry, and we will continue to pay attention to and explore corresponding technologies.
LiveVideoStack: What is the topic of your speech at this Shanghai conference? Why is this a topic?
Ma Qian: The theme of my speech this time is "Device-Cloud Integration Opens up a New Vision of On-Demand". Device-cloud integration is the core architecture of our Volcano Engine VOD. In the past three years, we have refined leading audio and video technology and video cloud high-availability service architecture relying on industry-leading products such as Douyin and Xigua Video. The optimization practice of business experience and cost has accumulated a wealth of platform capabilities and data decision-making power. We hope to share some of our experience and thinking in practice, and discuss with you.
LiveVideoStack: In the previous technology sharing, you mentioned that when the business scale effect reaches a certain stage, besides the experience, the cost is also crucial, and you put forward that the relationship between cost and experience is a "non-zero-sum game". The point of view stuck with me. So in the recent period, do you have any new thoughts and ideas on how to balance costs and complete a refined experience?
Ma Qian: When experience optimization enters the "deep water zone", cost optimization will become more important. Establishing a judgment on RoI is the core point of balancing cost and experience optimization. The cost optimization we are currently doing is mainly based on the necessity of business integrity. The difficulty we are currently facing and what we are continuing to work on is to gradually establish an understanding of the relationship between cost optimization and experience impact. With this breakthrough, there will be data science basis and higher efficiency guarantee for subsequent optimization.
LiveVideoStack: Different videos have different image quality problems, and it is difficult to fix them with a unified technical solution. How to "prescribe the right medicine"?
Ma Qian: From the perspective of visual perception, the following factors affect the image quality: clarity, color, fluency and picture defects. Our approach is to first analyze the quality problems of the video, and then generate corresponding personalized repair modules in each dimension, and arrange and optimize them.
For example, in the practice of fragment renovation, each fragment has a set of targeted restoration solutions, which is equivalent to "prescribing the right medicine" to achieve the most ideal restoration effect. Taking "Gourd Brothers" as an example, its screen mainly consists of two parts, the ink painting part and the paper-cut part. We use an algorithm to accurately identify and distinguish them, and then keep the ink part blurred, and repair and enhance the paper-cut part, so that the repair result is clear and the artistic style can be maintained.
LiveVideoStack: Where else can the technical space for experience optimization come from?
At present, we are trying to better understand the quantifiable scope of the technology space through QoE modeling. By associating business indicators, user perception and technical indicators, we can further plan the path of refinement and optimization, or the replacement plan of key indicators.
LiveVideoStack: How do you understand data-driven business growth? How much does the change of QoS affect QoE? To what extent should QoS be optimized to effectively improve QoE? So far, is there an accepted answer in the industry?
Ma Qian: QoE modeling, or indicator attribution is currently a difficult problem in the industry, and there is no standard answer. Our current practices may also be just the tip of the iceberg.
Experience is a very complex issue. It will be affected by many factors such as equipment, network, user behavior, and business scenarios. Even for the same user, the experience may change over time. How to gradually reduce this complex problem to understandable technical indicators requires a lot of experiments and data support. In our current practice, breakthroughs have been made in indicator attribution, trend demonstration, and boundary values of some scenarios, but more exploration is needed for refined scenario understanding.
LiveVideoStack: Up until now, when audio and video products are technically optimized, there are still soul torture problems that cannot be escaped. What do you think are still there? How does the Vulcanic Engine continue to innovate in technology in the face of these problems?
Ma Xi: A very direct question: " What are the benefits of this technology to the business, and how to evaluate it? "
In order to better help the business solve the appeal problem, or understand the actual value of our corresponding optimization. The VOD product of the Vulcan Engine focuses on the construction of two capabilities, the quality platform and the consultative service, and the platformization and standardization of data discovery → directional drive optimization → data closed-loop verification. More efficiently optimize pain points and track results.
LiveVideoStack: The vision and goal of the Vulcanic Engine is "responsible for the playback experience and the cost of use". Now, if you want to rate the current video service, how much would you rate it? What else needs to be worked on?
Marcy: Overall 8 points. From the perspective of on-demand commercialization, our multimedia SDK is currently doing well as a whole. It is basically at the top level in the industry in terms of stability, experience/cost strategy and scene optimization. In addition, we have provided a series of operational refinement products for VOD, such as the quality platform, which can enable the business to more intuitively perceive the status quo indicators, problems, and optimization trends.
In the future, we need to further understand ToB business needs, and strive to meet more product and scenario needs.
LiveVideoStack: Will Volcano Engine Video Cloud implement "cost reduction" as a long-term theme?
Ma Qian: Cost has always been one of the core competitiveness of cloud services. Compared with the traditional unit price optimization of resources, we are also good at reducing usage because we are a cloud service that evolved from a middle platform, and can help customers solve cost problems in multiple dimensions.
LiveVideoStack: Thank you for your answer and look forward to your wonderful speech at LiveVideoStackCon 2023 Shanghai.
▲Scan the QR code in the picture or click " Read the original text " ▲
Check out more exciting topics of LiveVideoStackCon 2023 Shanghai Station