This is a talk at Taipei Video Tech #5. The presentation illustrates how a Japanese streaming service handles chaotically supplied contents for 24/7 linear streaming. It also describes the background behind the fact that unique circumstances between the video streaming industry and broadcasting industry in Japan create the chaos.
Taipei Video Tech #5 talk : A Japanese Way to Maintain Constant Quality on Streaming Chaotically Supplied Contents
1. A Japanese Way to Maintain Constant Quality
on Streaming Chaotically Supplied Contents
Yusuke Goto, AbemaTV, Inc.
- 2019.12.27 Taipei Video Tech #5 -
2. Yusuke Goto
五藤 佑典
https://ygoto3.com/
@ygoto3_
● Majored in Graphic Design at California State University, San Bernardino
● Software engineer @ CyberAgent, Inc. and AbemaTV, Inc.
● AbemaTV Development Division
○ Lead New Device team
○ Lead Streaming Client team
○ A Developer Expert in video technology and product design
3. Everything about video tech in Japan is unique.
Unique practices, original specifications ...
4. Everything about video tech in Japan is unique.
Unique practices, original specifications ...
5. A Japanese video streaming service, AbemaTV.
Its workflow is also affected
by the unique practices and specifications.
8. The Company
Internet media
● Blogging platform
● Music streaming
● Game
● Curation
● Ad product
Development
● Ad agency
Broadcasting
● News
● Documentary
● Drama
● Anime
● Sports
● Music
● Movies
Internet TV station
15. 20 Channels * 24/7 + 18,000 Original Episodes
All Free
No need to sign up. Just visit and watch.
*premium features are additionally available
100% Professional
The contents are 100% produced by
professional experts in the industry.
Completely Programmed
Streamed 24/7 like broadcasting.
A Variety of Categories
Original live contents / drama / news /
anime / sports / etc.
16. We Do Everything E2E
● Studio work
○ Shooting
○ Editing
● Production
○ Pre-production
○ Post-production
● Mastering
● Development
○ The streaming system
○ The applications
○ The ad system
● Ad production
23. -24.0
LUFS
Linear Streaming and Server-Side Ad Insertion
CFR
Progressive
Interlaced
-18.0
LUFS
VFR
16:00 - 17:00
Slot A - Purchased
17:00 - 18:00
Slot B - Recorded
18:00 - 20:00
Slot C - Live
20:00 - 21:00
Slot D - Recorded
24. Example : AbemaTV’s Ad Regulation for Submission
Duration 15s/30s/60s/90s/120s/150s/180s/300s
Container Format MP4
Frame Size Full HD (1920 x 1080)
DAR 16:9
Video Bitrate (avg.) 20 Mbps
Video Encoding Mode 2-pass encoding
Scan Type Progressive scan
Video Codec H.264
Frame Rate 29.97 fps
Profile High Profile / Main Profile
Audio Channels Stereo 2ch
Audio Codec AAC-LC
Audio Bitrate 320 Kbps
Audio Bit depth 16 bits
Audio Sampling Freq 48 kHz
Audio Peak Level >= -3 dDBFS
Loudness (avg.) -18.0 LUFS ± 1LU
26. Sources Come from Everywhere
● Broadcasters
○ Their workflow is much simpler and
fixed
● Masters for BD/DVD
○ Too much info
● Old contents
○ Legacy codec
■ Can’t handle
w/ the latest software
● Sources originally for another purpose
○ Digital signage
○ Web video
■ Fewer framerate
■ VFR
○ Archives of game streaming video
○ Hardware-dependent codec
■ Via hardware-recording
● Multi-language sources
○ Multi-track
■ Primary tracks platform-dependent
● etc.
○ All the video sources that
are not optimized for streaming
31. Why MP4 ?
● Good for our needs
○ Small file size
○ Good compression for 4:2:0 chroma subsampling
■ Good enough as a de facto media format
○ Decodable everywhere
■ Playable everywhere
○ Fast transcoding
○ Simple structure
■ Easy to handle GOP
■ Easy to convert sources to I-frame-only data
32. Why MP4 ?
● MP4 already seems to be the standard for delivery
○ Major media solution providers primarily support MP4
■ AWS Elemental MediaLive
● mp4 only
■ Harmonic
● mov/mxf/mp4 - (mp4 is the stablest)
■ We do not use these solutions, but good to have other options
34. MP4 is hard to handle
● MP4 is an effectively compressed format
○ Easily succeeds negative factors from the original sources
■ Broken decode order
■ Wrong header info
Not all the content operators are video experts
38. A MAM solution is good, but ...
● Does not support all the media formats we receive
● MAM companies are usually not familiar with the Japanese manner
○ e.g., DALET is a French company
○ 30 fps / NTSC ? No, PAL does not use it
47. Loudness
-18.0 LUFS
● AbemaTV’s original specification
○ ARIB T032 specifies -24.0 LUFS
■ Not enough for mobile devices
○ Optimized for playback on mobile devices
○ Follows AES’ Recommendation for Loudness of Audio Streaming &
Network File Playback
■ -16.0 to -20.0 LUFS
51. Various Kinds of Platforms to Support
● iPhone
● iPad
● Apple TV
● Android Smartphone
● Android Tablet
● Android TV
● Fire Tablet
● Amazon Fire TV
● Google Chrome
● Mozilla Firefox
● Internet Explorer
● Microsoft Edge
● Apple Safari
● Google Chromecast
● Google Daydream View
● Amazon Echo
● LINE Clova Desk
● SONY BRAVIA
● Panasonic VIERA
● TOSHIBA REGZA
● Sharp AQUOS
● FUNAI
● LEOPALACE21
Life Stick
● CCCAIR Air Stick
● ...
62. Conclusion
● To keep constant quality on streaming chaotically supplied contents
○ Developed a unique and original specifications
○ Partly adopt parts of various standardized specifications
○ Developed tools and workflow to patch gaps