It’s National Novel Writing Month (NaNoWriMo) again. This year I wanted to participate, but do something a little bit different. Last year I branched out from the traditional novel by writing short stories. This time I’ve decided to produce a podcast in real time.

Real time podcast

What does that mean exactly? Each day I write an episode for the podcast, record the script, do all the audio engineering work to edit and clean up the quality, write show notes, format a transcript, prep the podcast files for update, and schedule it to release the following day. I decided to push live each day at 14:00UTC so it’s a convenient day time for both Europeans and Americans (sorry Australia).

I wrote one introduction episode on the eve of November to have something ready to go on day 1, but otherwise I am constraining myself to just writing every single day of November. Is it a smart thing to do? Certainly not. Is it easy? Hard no. Am I going to make it? Time will tell!

Solarpunk

What’s my podcast about? I’m glad you asked.

It’s about Solarpunk, a genre and movement that tries to envision a future where we have leveraged technology and infrastructure to become balance with our environments and within our communities. Want to know more? Check out the first episode.

My series discusses writing prompts in that setting, covering a single writing prompt per episode and diving into the inspirations and considerations for it. I offer some advice and try to motivate the writers.

The setup

This post isn’t about that, though. I was asked to share my setup and process and that’s what I’m going to do.

Let’s start with the hardware:

• Microphone - MXL 770 Cardioid Condenser
• Pop filter - Aokeo Professional
• Stand - Neewer Suspension Boom
• Cables - Amazon Basics
• USB interface - Focusrite Scarlett 2i2

Looks pretty fancy, doesn’t it? The mic is pretty solid but picks up the room quite well. It’s wonderful if you want to sit back and play the guitar, but unless you’re in a quiet space you’ll pick up the room acoustics and background noises. Unfortunately for me, I’m not in a quiet space. My office is a nook in the middle of our apartment hallway. I also have no sound insulation or absorption nearby. If you are looking for a really professional sound that stuff would probably matter to you. This reaches “good enough” quality for me, though.

Cleaning up the Focusrite Scarlett 2i2 on Linux

The Focusrite interface is important for me to be able to use this external hardware and have it actually work on Linux. There’s a lot out there that doesn’t. That being said, recording in via the 2i2 brings in the sound in a single channel. That means I can record in mono, but some things don’t work well with it.

So I remap the mono input to stereo in my pulseaudio config:

load-module module-remap-source master=alsa_input.usb-Focusrite_Scarlett_2i2_USB-00.analog-stereo master_channel_map=front-left,front-left channel_map=front-left,front-right

And while I’m in there, I also set up an echo cancellation sink:

load-module module-echo-cancel source_name=noechosource sink_name=noechosink
set-default-source noechosource
set-default-sink noechosink

Now that’s fancy!

I’ve read that this stuff would be easier in Jack, but I’m really bad at the audio configuration. I couldn’t get it to work at all. Pulseaudio gets the job done for now.

Alternatives

So if that sounded like too much work to be worth it, it probably is. It’s not like I have the rest of the skills or environment to get studio-quality audio anyway. I actually get pretty decent recordings out of this relatively cheap headset with mic:

• Headset - SENZER SG500

If you’re looking to duplicate my setup, I’d just do that instead.

Scriptwriting

So that’s it for the audio hardware setup. Now I have to write what I’m going to record. When I’m making video how-tos for Youtube/Peertube I don’t really bother with this step. I typically make a list of topics I want to cover and then just speak freely using those as a rough guide to make may way. This podcast requires a little more work. For one thing, the writing prompts I’m covering in my podcast aren’t my own. I’m working with a really smart group of people who came up with them and are helping me with links and comments for my episodes.

So I need to take these resources they’ve given me and run with them, turning it into about 10 minutes worth of content. Why 10 minutes? Well, it’s about the amount of time you can expect someone’s undivided attention (hence the format for TED Talks). It’s also about as much as I can reasonably write, record, and produce in a single day. Everyone wins! Besides, by keeping them short and punchy there’s really no excuse not to listen. You don’t need to set aside time for this thing. You can listen while you brush your teeth or something.

But writing needs to happen and to do that I use my handy editor vim and a plugin called vimwiki. It lets me quickly write in markdown format and have the full advantage of easy hyperlinking between content. I love my wiki life so much and I need to access things in it so often when I’m not at my computer that I actually rig it up to display on my website.

There’s a bit of custom code involved in that process, but mostly I’m just keeping the wiki files in Syncthing and have those accessible on my web server. A tiny bit of node converts markdown to HTML and I styled it to be pretty. Now I can get at those notes anywhere.

While working in Markdown, one of my favorite features is that single newlines in the source don’t matter. The content will wrap back up into a single paragraph when rendered to HTML. That means that I can have arbitrary line breaks. When writing that way I like to use a thing called Semantic Linefeeds, which basically mean I hit return when I finish the clause of a sentence. This is a great habit for prose writing which provides three key benefits:

1. Easy version control diffs (if you’re using version control)
2. Easy manipulation of sentences and clauses, especially in vim
3. Each line is broken in a natural way for reading it aloud

Writing is fun.

Recording

My recording setup is pretty straight forward. I’m using Audacity. There was some drama not too long back over telemetry and I think I was supposed to switch to a fork, but I forgot about that until I started typing this sentence, so that’s my excuse.

Actually, my first episode was recorded in a different way. I have OBS set up on this machine and it has some vocal filtering set up to clean up my sound when I do video recording or streaming. I thought I could save a step and just record myself there and pull the audio out of the video. It worked, but the noise gate in OBS clipped some of my words more than I liked. I switched to Audacity after that first episode.

Using my lovely hardware setup above, I prep my tmux into as many panes as I can to get the entire script on screen before I record. I don’t like trying to scroll or hit keys as I go. You can hear them, and I get lost. It’s easier if it’s just all there. If you hate trees, you could even print it out.

I try my best to read like I speak, except with less “ums” and “ahs”. Sometimes I need to remind myself to breathe more, or slow down, but the process comes to me pretty naturally. I’m not sure I have any advice here to give except Bill Murray’s advice about everything:

The more relaxed you are, the better you are at everything: the better you are with your loved ones, the better you are with your enemies, the better you are at your job, the better you are with yourself.

So relax, record, have some fun, smile, use emotion. If I screw up I just take a breath and record that line again and keep going.

Audio Engineering

I wanted to use that headline because it sounds so fancy. Ultimately what I’m doing at this stage is to quickly edit the audio until only the content I want to keep remains. I will select any notable breathing noises and CTRL-L them in Audacity to silence them. I’ll trim sections where I messed up. And… yeah, that’s it. That’s my entire editing process.

Now it’s time to make the audio suck a little bit less.

Correcting Voice Audio

Correcting Voice Audio

1. Noise Reduction

Step one, I grab a sample of empty space on my timeline when I’m not saying anything and take a noise profile from it. Then I select the entire track and run noise reduction using that profile. That cleans up a lot.

2. Compressor

If the audio has a lot of varying levels (louds and quiets) I’ll use a compressor to bring things a bit closer together. If my voice was relatively even then I’ll skip this step.

4. Use a high pass filter (HPF)

In audacity I do this and the following two steps in a single EQ filter:

First I do a high pass filter to drop sounds below 80Hz which is below my speaking tone. I also set a ramp up from 80Hz to about 125Hz, right where my change in clarity happens. If you’ve a very different voice from mine you might need to adjust these numbers a bit.

4. Boost in the mid-range

The main area of speech intelligibility is in the 1kHz to 4kHz range. I don’t need to do much here, but 4dB seems to have a remarkable increase in quality. I shape my increase like a bubble between 1 and 4 growing about 5db around 2,500Hz.

5. Add warmth to the vocal

To make the voice sound more warm and welcoming I boost a special range by 3dB between 160Hz to 300Hz.

6. Remove sibilance

I have a pretty strong lisp, especially as I get tired, so there’s only so much I can fix with my audio skills. Still, cleaning up a bit of the hiss helps. The best option here is to use a de-esser, but I don’t have one on this system. Instead I tweak my sounds between 5kHz and 7kHz down by 3dB. That’s where the S’s live.

7. Normalize

Finally I want to ensure my track peaks just below full volume and that everything isn’t too quiet. To do that I use a Normalize filter.

Add music

Finding music with rights you can use isn’t the easiest. For this podcast I was lucky enough to find an album on Bandcamp that’s Solarpunk themed and CC licensed. I’ve been picking one track from it for each episode.

My music arrangement pattern is really simple. I let the music play for about 10seconds before my voice comes in. I ramp the volume on the music back down again as soon as my voice comes in and keep it in the background for a minute or two. Then I fade it out to nothing slowly around the middle of the track.

Toward the end of the track I bring the second half of the song back in to that background level. At the end of my episode I say my thanks and mention the audio track artist and title. Just as soon as I start announcing that part I start ramping the volume back up. It should reach full about 2-3 seconds after my last word spoken, and then I leave the last 10 seconds of audio to play until the end of the episode.

There’s obviously a million ways to handle music arrangements, but this is something I can do automatically when it’s 3am and I’m exhausted and just want to be done recording.

Packaging it up

Once that’s all done I save down a wav and mp3 version of the episode and prep my podcast software with the latest episode. I’m using a self-hosted activity-pub enabled podcasting suite called Castapod. It was pretty simple to set up in Docker and it enables me to both create the podcast, its episodes, and easily arrange the setup of a bunch of podcast search services from Apple Music to Spotify to PocketCasts. It formats my RSS feeds, and also sends out announcements into the fediverse when my episodes release. People on the fediverse can follow that account, boost or like its messages, and reply, creating viewable comments on the website. It’s really quite cool.

I do a tiny bit more writing at this stage by creating episode notes. That’s usually a 2-3 sentence description of the episode, any links I mentioned in the audio, and a link to the song I used.

Unfortunately while I can add hashtags to the fediverse announcement posts for each episode, those hashtags aren’t enabled when the messages reach other servers. That means the podcast doesn’t show up under the relevant hashtags on Mastodon, for instance. I have to do a bit of boosting and mentions myself to get around that.

I put the audio files, transcript of the episode, and show notes into a cloud folder. The plan is to run through those transcripts to make subtitle files for those that need the text in an alternate way. In the meantime, people can find the full scripts on my wiki.

After November ends I hope to go back to the beginning and turn the audio episodes into video episodes on Youtube and Peertube. I may need a little break after this busy month, though.

Time will tell, right?