Twitch Renderer

I was interested in applying some Computer Vision techniques and playing around with Twitch streams. As Twitch chat is a big part of the experience, and especially the emotes are a very visual component, this code allows you to render streams in low quality at 30 fps and do the same for chat (instead of video capturing your browser). Due to latency concerns the irc data stream worker will download and keep an up to date database of Twitch emotes and badges (BTTV emotes too). Currently this database only downloads channel emotes when you ingest said stream as emotes from other channels are used “relatively” rarely in streams compared to its subscriber emotes and globals.

Dependencies:

• numpy
• zmq
• ffmepg

The API is really simple…

render everything

from twitchrender import Renderer

stream = Renderer('LIRIK')
for video, audio, chat in stream:
    # do stuff with frames

render video and chat only

from twitchrender import Renderer

stream = Renderer('LIRIK', audio=False)
for video, chat in stream:
    # do stuff with frames

Each Renderer instance can fetch 1 stream frame by frame. All you have to do is initialise the Renderer object and the overloaded __iter__ does the rest.

Renderer.__init__(self, channel, video=True, audio=True, chat=True, flush=True)

This is the renderer object used to mantain data-stream workers.

args:

• channel: the channel whose stream you wish to render
• video: bool which decides whether to render the stream’s video frames (default=True)
• audio: bool which decides whether to render the stream’s audio frames (default=True)
• chat: bool which decides whether to render the stream’s chat frames (default=True)

Renderer.__iter__(self)

This is the iterator used to yield data-stream frames. The data-stream sockets are flushed at each iteration, so at most this iterator will return results at 30 fps, but if this iterator is called less frequently only the most recent frame will be returned.

returns

• (360,640,3) video frame if video argument during __init__ was set to True
• (44100/30,2) audio frame if audio argument during __init__ was set to True
• (360,100,3) chat frame if chat argument during __init__ was set to True

NOTE: this is reliant on unsupported API calls and could break any time 4Head