Video and television

Encoding pipelines, live delivery, and the player applications that decode the result on a television. These are not separate jobs. A player and the manifest it reads are two ends of the same contract, and designing them apart is the usual reason an adaptive pipeline fails on the device it was built for.

01

Encoding and delivery

FFmpeg pipelines

Transcoding, scaling, and packaging. The interesting part is the ladder, because that is where adaptive playback either works or stalls every time bandwidth drops.

Adaptive bitrate

A set of renditions at matched resolutions with aligned keyframes, so the player can switch between them without stalling or drifting out of sync. Most pipelines fail here and the failure looks like a network problem.

Live delivery

Paths where a second of latency is a real cost. Audio pushed to an Icecast mount, and screen or browser capture relayed out over WebRTC with signalling that handles reconnection.

Player applications

Software for a screen in the living room, which behaves nothing like a phone. Shipped across five television platforms, with store submission for each.

02

Television platforms

Each store has its own review process, and each platform has its own idea of what a remote control is.

PlatformRuntimeWhat differs
Samsung TizenTizen, web stackStrictest store review of the five. Certificate expiry has to be handled, and Tizen Studio builds are their own toolchain.
Fire TVAndroidApp size limits and leanback behaviour. Review takes longer than the others and asks for the right declarations.
Android TVAndroidDevices with 2GB of RAM and no hardware decoder worth relying on. The baseline worth targeting.
VEGA OSVendor platformLeast documented of the group. Built against observed device behaviour rather than a published specification.
PhilipsVendor platformPer model and per region, which is where most of the testing time goes.
03

Work done in this area

VOD encoding with adaptive bitrate
FFmpeg pipeline producing a rendition ladder and packaging it for adaptive playback, so the client gets a variant set the player can move between. Called an excellent solution by the client.
Stack
FFmpeg, ABR, HLS, PHP, Python, Linux, shell, MySQL
Television player application suite
Player and catalogue shipped across Samsung Tizen, Fire TV, Android TV, VEGA OS, and Philips, with a store submission for each.
Stack
Tizen, Fire TV, Android TV, VEGA OS, Philips, HLS, ExoPlayer
Live audio to Icecast
A scripted path from source to an Icecast mount, built so it could be run repeatedly without a human watching it. Described as technically strong and well managed.
Stack
FFmpeg, Icecast, shell, Linux
Browser streaming over WebRTC
Browser capture relayed out over WebRTC including the signalling layer, with cloud and VNC infrastructure alongside it. Networking and systems work in the same project.
Stack
WebRTC, VNC, Docker, Linux, networking
Media infrastructure
The servers behind it. Nginx, the catalogue database, and the deployment, all containerised.
Stack
Nginx, Docker, MySQL, PHP, Linux
Media retrieval tooling
Download and retrieval tooling built to a stated scope and delivered quickly.
Stack
Python, Linux
Audio processing
Voice clarity enhancement and related signal work on recorded and live audio.
Stack
Audio processing, signal processing, Python, Linux
04

Questions

What decides whether an encoding pipeline works?

Two things, and neither is the codec. Keyframes have to line up across every rendition, or the player switches to a stream with no keyframe where it needs one and stalls for a whole segment. And the ladder has to match the source, because renditions below the original resolution give the player nowhere to go on a good connection.

Can you build an app for more than one television platform?

Yes, and that is usually the right way round. The player is most of the code, so the work per platform after the first is packaging, store submission, and device specific testing. Fire TV and Android TV share the most. See the contact page if you need more than one.

Do you work with existing pipelines?

Often, and that is usually the more useful job. A failing pipeline is easier to diagnose than a missing one, because the behaviour you are complaining about is measurable. Bring the ffmpeg command line and the manifest, and the problem is usually visible in one of them.

Do you handle live as well as on demand?

Both. Live work adds a constraint on demand does not: there is no second attempt. Anything that can go wrong has to fail in a way that recovers on its own, which usually means more code in the failure handling than in the happy path.