INXPO recently chatted with Streaming Media contributing editor and leading expert on live and on-demand video production, Jan Ozer. We got to ask Jan questions on his new book Video Encoding by the Numbers about eliminating the guess work from your streaming video.
How did you get into writing a book on compression?
I worked for Iterated Systems, a compression company, from 1991-1993. Since then, I’ve been benchmarking video compression technologies, starting with those focused on CD-ROM production, then DVD, then Blu-ray. In 1996, I started covering streaming technologies for PC Magazine, and two years later started writing for what ultimately became Streaming Media Magazine.
In the interim, I’ve reviewed most encoding-related products and services, compared codecs from Indeo to VP9 and HEVC, and developed a consulting practice that includes a variety of customers, large and small, that translate to significant real world experience.
This is my third book that focused solely on streaming video, my 20th (or so) overall. My last book was Producing Streaming Video for Multiple Screen Delivery, and its predecessor, Video Compression for Flash, Apple Devices and HTML5.
Tell us about your new book.
The book has roughly five sections. The first section provides the fundamentals on video and streaming production, defining terms and concepts. The second focuses on universal encoding decisions like choosing the optimal data rate, choosing bit rate control techniques like VBR and CBR, and setting I-frame, B-frame and reference frame settings.
The third section focuses on codec-specific decisions for H.264, HEVC and VP9. The fourth focuses on choosing and encoding adaptive bitrate technologies like HTTP Live Streaming and DASH. The last chapter defines per-title encoding, details why it’s important, and describes different approaches producers can take to implement it.
What’s different from other books about streaming production?
What’s new in this book is that all compression-related configuration recommendations are backed by both objective quality metrics and encoding-time considerations. I don’t just say use five reference frames, I show how the quality differs according to the different settings using eight test videos that represent a range of corporate and OTT uses, from screencams and Powerpoint-based videos, to real world and animated footage for movie distributors.
Integrating encoding time considerations into the equation yields some interesting results. For example, the chart above shows that using 16 reference frames adds very little quality to any of the videos over a single reference frame. However, using 16 reference frames almost doubles the encoding time. If you’re an encoding shop running at capacity, you can either buy a new encoder, or reduce the number of reference frames and push more files through your current setup with no visual effect on quality.
In short, the book is the first to base encoding recommendations on objective quality metrics, and by incorporating the encoding-time element, allows producers to make fact-based decisions on both configuration and capacity.
Who is this book for?
Well, it’s for the producer making both encoding and capacity decisions. It’s also for producers who want to learn how to use objective quality metrics, because the book details what they are and how to use them. as well.
All production-related chapters include instruction on how to configure the parameters discussed with FFmpeg, with the chapters on adaptive bitrate encoding also detailing how to use Apple tools like Media File Segmenter and Media Stream Validator, and MP4Box, which is a DASH segmenter. It’s the only book available that goes into that level of detail.
Compression is so fast moving, and there’s so much information available for free on the web. Why should streaming producers buy a book?
Most of what’s covered in the book isn’t available online, and probably will never be. I counted the number of MP4/WebM files created while I was writing this book, and it was over 5,000 files, all representing individual experiments and data points. There were over 1000 batch files relating to these experiments, both for FFmpeg and the quality benchmarking tools that I used. The data and lessons learned from these experiments isn’t out there for free, and never will be.
The book is also very up to date, with some entries from as late as January 2017. By covering topics like DASH, VP9, HEVC, and even the AV1 codec, I’m sure the book will be reasonably current through 2018 and beyond.
How and where is the book available?