As I’ve been wanting to start a podcast for the greater part of a year, I’ll be using the #30DaysofLearning at Lab651 to learn and document my journey of creating a podcast ( topic will be focused on Applied Artificial Intelligence ). I intend to document the steps that I’m going through to launch the podcast. Everything from the software, hardware, guest selection, the time needed for preparation, and general work that I will learn it takes to generate a podcast on Applied AI.

I’ve got a decent background knowledge of audio, technology, and streaming services in general and did a few podcasts called Real-Time Talk for some time at a publication called RT Insights back in 2016 and even earlier for an online radio startup I founded called Localtone Radio. I intend that I’ll give you, the reader, all the insights into what I run into, what hiccups I encountered along the way as I document this process, and ultimately I’ll have the podcast released in 30 days!

Step 1:

I’ve taken the approach of looking at the current landscape of podcasts that interest me. Here’s a short summary of those that I have been listening to. My interests are wide-ranging, but also still fit within what you might find in a general “geek” portfolio. Additionally, I’ve been trying to standardize on one Podcasting application as I believe most users do the same thing. I’m a tried and true Android user, so I have been using the Google Podcasts app as a way to find and explore new podcasts. I’ll cover the distribution stuff that I will learn later on, but in general, all podcasts that I have heard about from other sources, I’ve been able to easily find and add to the Google Podcasts app, which is good as I want all uses of various devices to be able to find the AppliedAI Podcast. 

Some podcasts I listen to and what I would like to model from them are the following:

Understanding that these podcasts listed above cover a national audience have sponsors and many episodes published these are a good example of something I will strive for, knowing full well that I just need to get out and get started and learn from the process.

Step 2:

AppliedAI Podcast

I’ve been reading articles by other podcasters that state you have a fairly unique name, but also something that is descriptive. As of now, I’m really drawn to the concept and focus of this being an “Applied AI” podcast. It fits well with the AppliedAI group that we formed, but it also shows my interest in how we take technology and apply it to the current problem that affects us today. As of now, I’m thinking of “Applied AI Today”.

Step 3: Description

One of the things you might notice when listening to a podcast is the host having a brief and concise message on what the podcast is about. In short, I think it’s really about what your mission is. Over the past few days I’ve been mulling over what this podcast is about and I’ve crafted the following as an introduction.

Intro:

“Welcome to the Applied AI Podcast where Justin Grammens and the team at Emerging Technologies North have conversations with experts and thought leaders in the fields of Artificial Intelligence and Deep Learning. In each episode, we cut through the hype and dive into how these technologies are being applied to real-world problems today. We hope that you find the conversations educational and applicable to your industry and don’t forget to connect with us to continue the conversation at AppliedAI.MN. Enjoy!”

I think this shows that I’m the host, but also that this podcast is a part of the larger Emerging Technologies North non-profit that I have co-founded and; leaves room for additional co-hosts, which I’m open to having joined me in the future. It also directs people off to the AppliedAI.MN group that is specifically related to the podcast and specifically is focused on how AI is being applied to real-world problems today. It’s amazing how over the next 3-5 years we are going to be seeing AI applied in some way to just about every occupation in the world.

Outro:

“You’ve listened to another episode of the Applied AI podcast. We hope you have enjoyed the conversation and are eager to learn more about applying Artificial Intelligence and Deep Learning within your organization. You can visit us at AppliedAI.MN to keep up to date on our events and connect with our amazing community. Please don’t hesitate to reach out to justin&AppliedAI.MN if you are interested in participating in a future episode. Thank you for listening!”

I think it’s important to at least tie off the podcast with an outro section and list ways in which the listener can contact you. My hope is that I can also use this podcast to find other interesting guests to be on the show! There’s a very large community of people in the Twin Cities specializing in AI and as of today, I’m still just scratching the surface. Additionally, with the podcast being such a broad topic such as AI, I envision there will be not only people sharing their experiences but also how they have learned the technology and of course a ton of up and coming experts that I have yet to connect with.

Step 3a: Recording the intro and outro

Great! Now I know what will be said during these parts, the work is needed to actually record it. While being somewhat musically inclined ( I played drums in a few local bands for more than a decade back in the ’90s and early/mid-2000s), I wouldn’t consider myself a songwriter or was eager to spend the time creating the music and voiceover for these parts of the episodes. Enter Upwork! I went online and posted the need for someone to take this script and create basic background music together and read these parts. Within a few hours, I got a number of applicants, and over the next day, close to 10 people applied to range from $50 to $200 for them to do this for me. They all seemed very capable and would do decent work. While the price is totally reasonable, I had an epiphany while I was talking to my Google Home. Why not just have the Google Home Voice AI do the text to speech? And on top of that, let’s have the AI write the music as well! There have been systems that have trained AI to write music for a number of years and while not perfect the Google Assistant voices have been getting better and better. Google even offers a cloud-based text to speech service that costs nothing until you reach millions of words. My decision was made. I went to work and created a script in Python that outputs an MP3 file. All you have to do is replace the “<text here!>” with the text you want it to say and it saves a file.

from google.cloud import text to speech




# Instantiates a client

client = texttospeech.TextToSpeechClient()




# Set the text input to be synthesized

synthesis_input = texttospeech.types.SynthesisInput(text="<text here!>")




# Build the voice request, select the language code ("en-US") and the ssml

# voice gender ("neutral")

voice = texttospeech.types.VoiceSelectionParams(

    language_code='en-AU',

    name='en-AU-Wavenet-C',

    ssml_gender=texttospeech.enums.SsmlVoiceGender.FEMALE)




# Select the type of audio file you want returned

audio_config = texttospeech.types.AudioConfig(

    speaking_rate=.9,

    audio_encoding=texttospeech.enums.AudioEncoding.MP3)




# Perform the text-to-speech request on the text input with the selected

# voice parameters and audio file type

response = client.synthesize_speech(synthesis_input, voice, audio_config)




# The response's audio_content is binary.

with open('output2.mp3', 'wb') as out:

    # Write the response to the output file.

    out.write(response.audio_content)

    print('Audio content written to file "output2.mp3"')


What’s really cool is that there are a number of languages and language_codes that it can output. Not only can it do more than 33 languages, but it can also do a number of accents within each language. For example, within English, there’s the US, UK, and AU. In the code above, I’ve chosen one of the AU ones for my voiceover. A full list of all options can be found in the Google Text-to-Speech docs.

Step 3b: Into and Outro Music

After listening to the voice that is outputted, I looked around for some music created by an AI that I thought would be mellow and set the tone for the conversion. Something a little airy and spacy is what I am going for. There’s an interesting project called MuseNet that I’m currently exploring and adding the voice-over over the music using Garage band. Pretty straightforward and while I can do this after I get the first podcast recorded, I have a path forward and am going to move forward with the next steps.

Step 4: Ask questions

I’ve had the pleasure of talking with a few friends who have started podcasts and they have given me some good advice. To name a few and their podcasts.

  • Dave Mathias
  • Dan Yaramoulk
  • Kurt Schmidt
  • Scott Allen

As these people have recorded many more episode that I have, some of them are greater than a hundred, here are a few of the general themes I’ve heard

  1. Have fun with it – I couldn’t agree more. Like starting a business, if you are in it for the wrong reasons, your likelihood of failure is much higher. Once it stops becoming fun, I’ll want to rethink if it’s something that I’ll want to continue to do.
  2. Just do it – Rather than waiting for the “right time” with professional equipment, world-class guests or interviews (heck, I don’t even know if I’ll be a good interviewer – I might suck at it). It’s better to just get going and iterate and improve over time.
  3. Automate as much as you can – The editing process of the raw audio can be very time-consuming. You might consider finding someone on Upwork to do the audio cleanup for you. As paying someone to do the audio cleanup is something I’ve never done before this is an area where I’ll be learning, but can absolutely see the benefit of saving my time for the recording and distribution and someone else to do the cleanup.
  4. Keeping it as a conversation  – I’ve been told that when you are interviewing someone in this context, there are times where the person who is being interviewed gets into a zone of conversation that’s almost magical. They become so engrossed in the conversation that you as the interviewer just need to step back let them take the spotlight and guide the conversation
  5. Plan in advance, but not too much – You’ll want to make sure you set aside time to do some upfront planning of the conversation. Not that it all needs to be planned out, but some general ideas on where the conversation will go. It might be wise to ask the person you are interviewing if they have specific areas in which they would like to discuss.

These are all great tips and have been on the other side of this experience ( being interviewed on podcasts myself ), my biggest fear is the unknown of being the interviewer. I’m definitely a curious soul and love to dive in and learn new topics, but being in the seat of an interviewer is going to be a new experience. As I got into these first few episodes, I’m going to try and be cognizant of keeping this to a conversational level and just try and roll with the way things go.

Step 5: TEST, TEST, TEST!

There’s a wealth of platforms out there when it comes to podcasting software. What I’ve mainly seen is that it breaks down into 2 categories: Recording and Distribution.

Recording

The tools and platforms have changed a lot since my early days of podcasting on Localtone Radio where I was interviewing bands and talking with them about their music back in 2007. In those days, I actually had built a system using the Open Source PBX System called Asterisk. It was actually a lot of fun. People could call a phone number that would route into my Asterisk server and I could push *6 during the call and it would start recording the conversation of each person on the call. It worked pretty slickly for the cost of a VOIP line back in the day.

I tested out a few different offerings:

  1. Zencastr – Browser-based tool that both myself and the guests can connect and record the session. It records separate tracks that the host can download and then mix themselves. They do offer a service that will mix the tracks, take out spikes in volume, etc. It’s an additional paid feature, but something that could be worth the time and energy. They offer unlimited time and users per month in their basic plan, which is a perk.
  2. SquadCast – Browser-based recording tool that shows both video and audio. It only records the audio, however (which is just fine). You can download separate files for each person speaking ( .wav ) or a file that is already mixed ( .mp3 ). Squadcast offers 5 hours of recording per month on their most popular plan, so not completely unlimited like Zencastr, but I think 5 hours a month initially will be more than enough to cover the frequency at which I’ll be posting.

If you read the podcasting media sites, it basically comes down to personal preference. Both of them record the audio locally and then send it to the server (as to avoid internet hiccups in real-time) and both have roughly the same features and abilities. 

As I personally enjoy having conversations with people and being able to see their faces and read their body language, I think this a great feature and I’m planning as of now, to start out with SquadCast and see how that works. I would welcome any feedback or comments on the recording platforms you have used!

Distribution

Like with recording options, there seems to be a whole slew of options when it comes to the distribution of the podcast. At the end of the day, I would like this podcast to be found on all the major platforms that at least I know of ( Apple, Spotify, Google ) and easily found. Additionally, I’m going to want an easy drop-in player for the website so people can listen through the browser. The two that I’ve been told about are:

  1. BuzzSprout – Seems to be low cost and pretty easy to use. They offer a free tier that has up to 2 hours a month of upload and a low cost (ad-free) plan at $12/mo.
  2. SoundCloud – They also offer a free tier and hosting of up to 3 hours a month and pro features for $12/mo.

As of right now, my focus is to just get a few episodes of the podcast done and will figure out distribution when the time comes. Like with the recording tools, I would love to hear if you have any feedback on the distribution platforms that you have used.

Finally, the other aspect of discretion that I’m trying to think through is whether each of these episodes is both shared on the Lab651 blog, the AppliedAI.MN page, which the website still needs to be created and likely won’t be done by the time I get these episodes done. At the end of the day, it would be good to centralize the location where Google and other search engines find these, even if that might change someday.

As I conclude this initial post, it’s been a lot of fun working on getting this to a point where I have done the planning to create a description, equipment, software, distribution, and some rough topics. I the next article, I will talk about how the recording, mixing, and editing went and what I have learned during this #30DaysofLearning!

~Justin Grammens