Episode0.3 was very interesting indeed, and whilst it fixed many problems, it also introduced some others too. I had previously had a problem with xvidcap, which was my main screen capturing software. During the filming of e0.2 xvidcap started to experience problems doing simple recording. It would crash out without a real reason. I got incredibly frustrated sitting there, talking into a microphone for three minutes, before realising that there was no point in me continuing, because the screencast had stopped.
So I did some tests with gtk-recordmydesktop and I must say it worked excellently. The ogg files it generated were sharp, clear and small in size, which was perfect. The video footage of me was captured with the Olympus C-8080 and the audio on the Nokia N810. I created the overlays for the upcoming segments in Inkscape as transparent png files, intending to fade them in with a cloud luma effect in Kdenlive.
The spinning cube/box seen in the intro and in the “What’s on today” section, was created in Blender and then rendered out as a RAW AVI file, to be imported into Kdenlive. The process was simple, I created five planes, the first four I set to have animated textures, a la the video footage, and the last was the floor, which was set to have a receive shadows only texture. The animation was done with simple keyframing and the background of the world was set to white so that the floor plane was indistinguishable.
I arrange my files quite tidily now for each episode. I have the following folders;
rawvideo : all the raw footage from the camera, and screencasts
encvideo : videos that have needed encoding to make them usable
rawaudio : all the audio from the N810, from both footage and screencasts
encaudio : the encoded/processed audio
overlays : the png overlay files
images : any other images that need to be displayed
songs : the song files for the episode
segments : any rendered segments, like intro, or segswitch
out : the main render directory for complete episode renders
Audio processing was done with Audacity, and takes a fair amount of time, I’d be interested to know if anyone else can batch these edits. I do the following;
- Take the raw 8Khz audio and push it up to a project rate of 22Khz
- Use an equaliser curve to bring up the bass and make it sound like I’m there ;)
- Amplify to normalised level
- Noise removal on the entire track
- Save in the encaudio folder
The main reason for the 22Khz upsample is the fact that Audacity crashes if you try to do noise removal on 8Khz sounds. It took a good half an hour, to an hour to go through and edit up these files.
So I started beginning to edit together the footage in Kdenlive, or at least, that’s what my plan was. I never quite got there. Kdenlive started crashing out randomly and lead me on a two week voyage which ended up where I started. In short, I had my 30 minutes of footage, I had taken 4 hours to get there, and there wasn’t a damn thing I could do with it.
Enter Cinelerra. After months of shunning the video app, I decided that it seemed like the only one available on Linux that could actually do what I wanted. So I set to work, initially producing test files and seeing how they worked. In fact, the first bit of footage to be created was indeed the intro sequence. This was rendered using Cinelerra’s internal renderer to an ogg file and stashed in a folder. As was the segment switch part.
Then I began to bring in video and audio segments. In actual fact the syncing of audio to video was very easy to do in Cinelerra. Each video brought in the audio the camera had recorded, apart from screencasts, and the wave form was displayed on the screen. All I had to do was align the waveform of the file from the encsound folder with the waveform from the MOV video, and I was laughing. Then I just deleted the MOV audio track, and off I went.
The intro required something special, namely the small version of the spinning box. This was achieved using Cinelerra’s projector. I used the shift drag in the compositor window to zoom out on that video track, and hence make it smaller. Then I just positioned it, and it looks pretty good.
The final project was then exported out to a Quicktimeforlinux format using the YUV Planar encoding, which for 27 minutes at 29.97fps, using a resolution of 640x480, took up a staggering 11Gb of storage. This was then used to create the AVI XViD files, which were encoded using ffmpeg via the winff front end. The ogg file was directly rendered from Cinelerra. One thing which amazed me this time round was the quality of a) the video and b) the sync between audio and video. I think it comes mainly from rendering an 11Gb flat uncompressed file to begin with and then re encoding it later. The sound was a little out of sync with the video as it progressed, but as we were only recording short segments, it was easy to nudge it back in again.
The lighting was replicated from episode0.2 and turned out to be pretty good, except the video segments of me, looked rather dark, so I used a plugin in Cinelerra to brighten them up a little. This seemed to have no effect on render speed, or play back speed showing again how good Cinelerra is at doing its job.
Episode0.4 will be the first episode to actually use a real video camera. Unfortunately it doesn’t have an audio in, and so I will still probably have to do the whole rigmarole of having the N810 be the audio recorder. It will be interesting to see how the video will change now that we’re recording in a higher quality, and in mpeg2 rather than MOV format. It’ll also be interesting to see how the sync between audio and video works out, now that I’ll be using a proper video device to record the segments with.
I must admit, I enjoy the new challenges that come with each new episode, apart from the ones that I can do nothing about. Working out how to do a tricky transition is cool. Trying to fix a segfault is not, at least not when I have such a tight deadline.
Well that’s it for the production notes for episode0.3
Stay tuned for more ;)