按标题视频编码:现在开始的时候了

Let me get straight to the point. Adding a new codec to your encoding mix, whether it’s HEVC, VP9, or AV1, is expensive, time-consuming, and somewhat risky. 将逐标题编码添加到编码工作流中可以提供许多相同的好处, but should be much less expensive, much simpler, and much less risky. 虽然所有公司都应该带着某种程度的恐惧来处理编解码器的变化, now is the time to embrace per-title encoding, and it’s never been more widely available.

Let me explain. 从流媒体的角度来看，NAB 2018将因AV1的首次亮相而被铭记. However, with that debut came some harsh realities, like an encoding cost of at least 10 times more than VP9, 还有2年的延迟，直到解码硬件出现在设备上. Although Facebook, Netflix, and You-Tube have all begun AV1 trials, if your streams aren’t viewed by millions, then AV1 makes little sense, at least in the relative short term.

与此同时，受苹果加入HLS的鼓舞，HEVC才刚刚开始看到一些发展. Although the benefits are clear, the costs of HEVC are all additive, 包括在执行前进行重要的播放器开发和兼容性测试，以及按下Go按钮后增加的编码和存储成本.

In contrast, 每个标题编码通过为每个标题创建优化的编码阶梯，使当前编解码器更高效. 这些技术评估源素材的复杂性, and can adjust the data rate, the number of streams in the adaptive group, and even their resolution.

Netflix在2015年底推出了按标题计费的模式，YouTube在2016年初宣布了这一做法. The first commercial implementation came in Capella Systems’ Cambria encoder in mid-2016, and now virtually every vendor offers this feature, including cloud vendors Bitmovin, Brightcove, and Mux; encoding vendors AWS Elemental, Harmonic, and NTT Electronics; and optimization vendors Beamr, EuclidIQ, and ZPEG.

每个标题的好处类似于高级编解码器——更高效的流媒体和更高的QoE. For example, 在分析他们新的“基于镜头的”编码结果时, which encodes each scene in a video separately, Netflix reported this in a March 9 post 在其技术博客上:“对于低带宽连接的会员, 我们将以相同(甚至更低)的比特率提供更高质量的视频. 对于高带宽连接的会员，我们将以更低的比特率提供同样的高质量. 当网络吞吐量急剧下降时，许多成员将经历更少的重缓冲和质量下降.”

一个有趣的技术注意事项是标题编码之间的区别, video optimization, and shot-based encoding. When it debuted, Netflix使用其标题编码分析作为评估整个视频编码质量的一种方法, and then encoded using two-pass VBR. This is how Cambria’s per-title feature works; the system gauges encoding complexity and then can encode using constant bitrate (CBR) or variable bitrate (VBR).

相反，优化技术是逐帧工作的. I’m most familiar with capped CRF and ZPEG, 根据场景的复杂程度来改变编码视频文件的数据速率, 但比姆和欧几里得的优化技术工作原理大致相同. Here, you can’t use a traditional bitrate control technique; in essence, 按标题编码技术是位率控制技术.

如果您仍然在使用CBR，因为您认为它提供了最耐缓冲的体验或最广泛的播放兼容性, 那么你可能更喜欢传统的逐标题技术, 至少适用于自适应比特率(ABR)流媒体，而不是可下载的视频. 在考虑任何技术时，您的关键问题是，“我可以应用CBR吗??” If you’re using 200 percent constrained VBR, 那么你可能就不会从每个标题或优化中看到太多额外的流变化.

基于镜头的编码显然是未来的趋势，因为它在打破许多既定编码惯例(如固定GOP大小和比特率控制技术)的同时呈现出许多优势. 为什么不使用优化的编码参数对每个场景进行编码, rather than applying one schema to the entire video? 为什么不在每个场景的开始都放一个i帧呢, rather than every 2 or 3 seconds? Why use a 2-second GOP length for a 3-second scene?

To my knowledge, although shot-based encoding is being worked on, it’s not available from any encoding vendor today. Don’t wait for it, though. Per-title or optimization is here now, 并且应该被列在2018年所有编码专业人员的待办事项清单上.

[This article appears in the June 2018 issue of Streaming Media magazine as "Per-Title Encoding: The Time Is Now."]