Using Riverside.fm for AI Summarised Interviews

Riverside is my favourite video interview recording platform, I use it for personal projects, client briefing sessions, in-depth interviews, podcast clients and basically anything where I don’t want to have to stop and make notes along the way, but I know those notes, summaries and interview content will be essential.

In this short sponsored article I’ll share my workflow for recording interviews, downloading Riverside’s auto-generated transcript and using online AI tools to create summaries, detailed notes and draw out the best quotes and moments from each interview.

Why I use Riverside and you should too?

riverside.fm discount code 30%

I’ve gone into a lot more depth on a whole range of different options when it comes to recording ‘Zoom’ interviews in this previous post, but Riverside is my go-to choice for these reasons:

  • Pristine video recording – Each video feed is recorded locally and uploaded to your Riverside studio so they are not affected by internet drop out or live encoding errors.
  • Individual video/audio stems – Each source, video, audio, shared screens, are recorded separately for later editing, these are recorded in the source resolution, up to 4K.
  • HD screen Recording – If anyone shares a screen this can be recorded in HD and synchronised to the video content
  • Auto-sync and Premiere Timeline Download – Everything is synchronised and available to download as a prepared Adobe Premiere Pro project and timeline.
  • Auto Transcribed – Download the conversational transcript as a .txt file or .srt subtitle file.
  • Ease of use – Just send a link and open with Chrome. That’s all the technical hurdles anyone has to clear! Perfect for even the least tech-savvy user.

Exclusive Riverside Discount Code!

Save 20% on any Riverside.fm membership plan with the discount code: jonnyelwyn

Looking for even more AI Post Production Tools?

Rummage through my regularly updated curation of the very best AI tools for post production professionals, including Riverside!

Riverside’s New AI Magic Features

I recently used Riverside to record this in-depth interview with BAFTA winning Editor Simon Smith, breaking down the iconic One Way Out moment from one of his episodes of hit TV show Andor.

Riverside’s native screen-recording was hugely important to make this interview work, given that we need to see Simon’s timeline to understand what he’s referring to, shot by shot.

Riverside have recently added some new AI-powered features to generate social-media ready clips with automated captions and silence removal.

10 x Riverside AI Magic Clips

I tested these AI-generated social clips on my interview with Simon and the results are pretty good. It chose some decent topics and the subsequent clips were fairly accurate too.

It would be nice if you could prompt it to find a clip on a specific topic, but my workaround in the next section will allow you to do that instead.

Editing these in Riverside was pretty intuitive, being able to toggle on/off input sources, adjust the layout of them within the clip (although I couldn’t see how to adjust the framing of each source?) and using the transcript to adjust what was included in the clip.

What’s this? A timeline for ants?!

One thing that was a little tricky to use was the timeline bar at the bottom as there is no ‘zoom control’. So my tiny 1 minute social clip, selected from a 90 minute source file was impossible to adjust with any fidelity at such a ‘zoomed out’ view.

I also know that some further AI powered tools to help creators get results quicker are in the works too, so keep a close eye on their site.

I genuinely love using Riverside and have recommended it to almost everyone who joins me on a call. Save 20% on any Riverside.fm membership plan with the discount code: jonnyelwyn

How to use AI to summarise interviews, client meetings and more

If you’ve not yet recorded the remote video interview then I’d obviously recommend using Riverside to do so, you even get 2 hours of (watermarked) recording time when you sign up for free.

If you already have a source audio or video file you want to work with, maybe a complete edit, existing video or YouTube link, you can either use one of the many AI transcription services to generate a transcript to work with or use a Premiere Pro .srt file for more accurate search and verifying the veracity of what the AI is telling you. (More on this at the end of this post!)

My AI-Powered Riverside Workflow

Step 1 – Record in Riverside

Record your interview in Riverside. This is very easy, just create a new studio, set a couple of essential settings (explained here) and share a link to your interviewee.

All they need to do is open the link in Google Chrome and you’re all set. Ideally, both sides would wear a pair of headphones, but Riverside’s noise removal and vocal isolation has always worked well for me.

Step 2 – Download Riverside Auto-generated Transcript

Once finished Riverside will auto-transcribe the interview, which you can download as either a conversational transcript as a .txt file or a subtitle .srt file.

The transcript is great for pulling quotes or entire paragraphs of text from, while the .srt file is ideal for subtitles and maintaining the timecode metadata in subsequent operations.

Step 3 – Use Claude.ai or Chat-GPT

Open your AI bot of choice and prompt the AI with how you want the LLM to act, the context of the interview, the nature of the kind of things you want it to do and the details you want it to provide.

Then paste in your transcript or upload the file, depending on what your AI supports.

As it is early days for these large language models, I will often experiment with both Claude.ai and Chat-GPT to see how the results compare. In both cases these have free to use tiers.

Depending on the length of your file, the free version of Chat-GPT may not be able to process it, in which case use Claude instead.

I have tried telling the AI-bot that it is an expert in a certain area, or that I’m trying to ask questions like a certain role (producer, product manager, researcher etc. ) but it doesn’t seem to make a huge difference to the results.

e.g.

You are a highly experienced video editor with a strong storytelling sensibility and a keen eye for detail. You have a lot of knowledge of film and TV post production workflows, techniques and pipelines. I'm going to give you a transcript of a conversation between two video editors and I want you to summarise it in two paragraphs and then list 10 of the most common topics with specific quotes illustrating each point:

or

Here is an SRT file converted to a .txt file. It has a start timecode, end timecode and the subtitle. It is a conversation between two editors. I want you to summarise the conversation and then extract 10 of the most interesting topics with quotes for each.

With each quote provide the starting and ending timecode for each quote, even if it spans multiple subtitles.

Step 4 – Ask lots of questions

Now comes the fun part! Ask lots of questions of your transcript and getting the AI to do some heavy lifting. What kind of things can you ask it?

Depending on the content, I ask for:

  • A two-paragraph summary
  • Bullet points of key topics with quotes for each and timecodes
  • Follow-up questions asking for more examples about a specific topic it identifies
  • (Client briefing) – Summary of the questions they asked
  • To-do lists
  • Check lists of technical processes described
  • Finding specific topics and summarising that portion of conversation
  • Timestamped list of major topics (useful for YouTube chapters!)

Ask for Feedback to Improve Your Questions

Sometimes I will ask it for feedback on the questions I’ve asked, to help me consistently improve over time. This could be especially helpful when recording client-briefings or in-depth interviews.

Never miss an opportunity to get feedback no one else will give you!

Here is some feedback on the interviewer's questions, with suggestions for how they could be more open-ended:

Overall, the interviewer asked good questions that prompted detailed responses from the editor. However, some questions could have been phrased in a more open-ended way to elicit even richer responses.

Asking "How" instead of "Why" - Example:

At 00:22:02:22 the interviewer asks: "Why are you holding it off?" This results in a shorter answer about the editor's intent. Asking "How did you decide when to use that intense close-up shot?" would have encouraged more detail about the thought process.

Using "What" instead of "Did" - Example:

At 00:09:27:05 the interviewer asks: "Did you kind of feel like, oh, I need this?" Using "What made you realize you needed those additional shots?" keeps the question open-ended.

Following up on interesting tangents - Example:

At 00:40:57:15 the editor refers to "blink thing is like when people are thinking." The interviewer could have asked "What made you start thinking about blinking in that way?" to explore this perspective further.

Asking for examples - Example:

At 00:19:31:19 when discussing tracking progress, the interviewer could have asked "Can you provide a specific example of when tracking progress really paid off?" to get a more detailed anecdote.

Using "Tell me more..." - Example:

At 00:27:31:09 when the editor references scene cards, the interviewer could have asked "Tell me more about how you use those scene cards in your process." This invites expanded explanation.

Overall, the interviewer had strong rapport with the editor and elicited lots of great insights. A few tweaks to wording could have encouraged even richer responses in some areas. Asking more "How" and "What" questions, following up on tangents, and asking for examples are techniques that can yield more detailed anecdotes.

Sometimes Weird Stuff Happens

Weirdly, if you prompt the AI in the first message saying “I am going to give you a transcript” and then intended to upload/paste one in the next message, it might hallucinate a response based on some ‘hypothetical conversation’. Claude gave me a whole response about editing a documentary about homeless people.

Here is a two paragraph summary of the conversation between the video editors, followed by a list of 10 common topics with illustrative quotes:

The editors discussed the challenges of working on a documentary film about homelessness. They talked about the difficulties of capturing authentic footage while being sensitive to the subjects, and the complexity of weaving together many disjointed stories into a cohesive narrative. Overall, they reflected on the emotional toll of working on such heavy subject matter and the responsibility they felt to honor the humanity of the people featured.

Throughout the conversation, several key themes emerged. The editors spoke extensively about story structure, discussing how to present the information in a compelling yet honest way. They also covered technical considerations such as lighting, camerawork, and editing techniques for interviews. Additionally, they touched on licensing concerns, rights issues, and working with a limited budget. Other topics included finding the right tone, building trust with subjects, and managing their own mental health while working on the project.

Topics:

Story structure: "We really need to think about the overall narrative arc and make sure we're presenting these stories in a way that flows logically."

Lighting: "Lighting these interviews is tricky because we're working in uncontrolled environments."

Licensing and rights: "We have to be so careful about only using footage we have the rights to."

Budget limitations: "This is an indie doc, so we have to work within a pretty tight budget."

Tone: "It's a fine line we're walking between depressing people and inspiring them to take action."

Building trust: "These people are opening up about very personal experiences. We need to establish trust."

Camerawork: "For these street scenes, handheld will give a sense of immediacy."

Interview editing: "Let's really take our time with cutting the interviews. They're the heart of the film."

Mental health: "This subject matter takes an emotional toll. We all need to take care of ourselves."

Honoring humanity: "Above all we need to honor the dignity of the people who shared their stories."

Sometimes it makes stuff up

Building scene dynamics
"I wanted to build to that close up. Let's drop that in. So that development of him here is almost, you know, nothing to do with or less to do with what he's saying." (00:24:18:22 - 00:24:26:04)

In this example the quote is actually more of a summary of what was said, rather than a verbatim quote, despite it giving me start and end timecodes.

So always cross-check that it’s accurate!

Bonus – Convert .SRT files to .TXT, .CSV and many other files

One of the problems with using Claude.ai to upload a .SRT file is that the format is not supported as an upload, but Claude can work with the .srt internal formatting, but only as .txt or .csv etc file. Or just pasting the content in.

EditingTools.io has a huge collection of super-useful post production tools, including a SUPERB online subtitle conversion tool that will quickly give you a file Claude can work with, and for free!

The huge benefit to using an .srt file instead of the transcript is the line by line accuracy of finding the right content, using the timecode stamps as the reference point.

This also makes it much easier to check that the AI tool has actually gotten it right.

The tool can also tidy up your subtitles, make them align to edit points when combined with an EDL or XML of your timeline and a whole lot more!

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.