How to generate podcast transcripts
Fireflies.ai creates accurate AI transcripts with speaker recognition, for just $10 / month.
Every couple of months, we test the best options for podcast transcription.
As of March 2023, we recommend Fireflies.ai.
Fireflies is designed for transcribing meeting notes, but it works well for podcasts too. It includes very accurate speaker recognition and a nice interface to edit your transcript while listening to the audio. If you take the Pro plan ($10 / month), it'll also generate a summary of your discussion.
The service has some minor issues: it sometimes hears additional speakers that aren’t there, and may have some strange sentence breaks. Overall, it's much better than anything else we've tested.
How do I use it?
Create an account (or log in with a Google account). Go for the Pro plan—$10 per month, with a free trial period.
- Upload your audio or video file.
- Wait 10-15 minutes for your file to process (for a one-hour recording).
- Click on the processed file to generate the transcript.
- Set speaker names and edit your transcription as needed (click “Save” after making your changes).
- Download your transcript and choose your export format (.docx, .pdf, .srt, .csv, or .json) formats.
In Appendix 2, you’ll find an un-edited transcription of an interview with Steven Pinker.
Let us know how you get on
How are you creating your transcripts? We would love to hear how you get on. Write to firstname.lastname@example.org.
P.S. We can create and publish transcripts on your behalf
TYPE III AUDIO can create, edit and proof-read transcriptions on your behalf.
We’re accepting selected clients to try our beta service.
Your first hour of transcription is free. After that, you’ll pay $100 per hour of audio. You’ll receive your ready-to-publish transcript within 2 business days. We can even publish it for you.
Write to email@example.com.
Appendix 1. Other transcription services
Below is a roundup of some other promising options (both AI and human-generated).
AI transcription services
As of February 2023, we think the best alternative to OpenAI Whisper is Azure Speech-to-Text. The transcript quality is similar or slightly better, and it has a good interface for reviewing the transcript while listening to the original audio. However, the service requires an absurdly specific audio file format, which you'll probably have to generate especially for this purpose (WAV; 16kHz or 8kZh; mono PCM). It also doesn't support speaker recognition, and creating an Azure account is surprisingly difficult (estimate: 10-30 minutes).
We previously recommended Verbalize, which uses OpenAI Whisper, is highly accurate, and includes speaker recognition. However, the service has been sporadic and glitchy, so we haven’t been able to rely on it. It also requires a subscription after a limited number of free uses.
Amazon Transcribe offers speaker recognition, but its output is in JSON format (?!), and the interface has a steep learning curve. If you've not used AWS before you'll have to configure S3 buckets and complete several other confusing tasks before you can get started (estimate: 30-90 minutes), and then you'll need to be comfortable converting JSON into a human-readable format. The transcript quality is similar to Whisper.
We haven’t had a chance to try the services below:
- The online version of Microsoft Word has a transcription option for up to 5 hours a month for those with a premium account (starting at $7 a month)—see this blog post for an overview of its accuracy.
- Otter.ai is widely used, and has both free and paid options. The previously mentioned blog post also reviews its accuracy.
- Google Cloud Speech-to-Text provides 60 minutes of free transcription per month, plus $300 in credits for new customers.
If you try any of these services, please let us know how you get on.
We'll continue trialling services over the next months. To get notified if our recommendation changes, click "Get updates" on the bottom right of this screen.
Human transcription services
Human transcription services are often sold as more accurate than their AI counterparts. However, when we compared OpenAI Whisper to Rev.com (a human transcription service) we found very little difference in accuracy.
The main benefits we see from human services are:
- Removing some filler words and false starts.
- More natural punctuation that improves readability a bit.
- Sometimes better at understanding different accents.
However, you’ll have to wait hours or even days for your transcript, and the prices often vary according to turnaround time (ranging from $50 to $200 per hour of audio).
Most services will offer a free demo for you to assess their quality—here are a few you may want to try out:
Appendix 2. Steven Pinker interviewed by Joseph Walker (first 10 minutes)
This episode is sponsored by Givewell. Imagine if every year you saved a person's life. One year, you rescued someone from a burning building. The next year, you saved someone from drowning. The year after that, you're out for dinner with your partner. Or maybe you're on a date and you notice someone across the room having a heart attack.
You perform CPR and save their life.
You would really be a hero. The truth is, we have an opportunity to do this every single year of our lives just by targeting our donations to the most effective charities in the world. How is this possible? Three premises number one, if you're listening to this podcast, chances are you make more than 19 and a half thousand US. Dollars per year post tax and are therefore in the richest 10% of the world. Number two, we can do 100 times more good for others than for ourselves by focusing on the parts of the world most in need. Because a doubling of income will always increase subjective wellbeing by the same amount. And three, in the same ways the success of forprofit companies isn't normally distributed, some charities are vastly more effective than others. But how do you find the most effective charities?
Well, since 2010, over 100,000 donors have used Givewell to donate more than $1 billion. Rigorous evidence suggests that these donations will save over 150,000 lives and improve the lives of millions more. Here's how. Givewell spends over 30,000 hours each year researching charitable organizations and only directs funding to a few of the highest impact evidencebacked opportunities they've found. Here's the best part using Givewell's research is free. Givewell wants as many donors as possible to make informed decisions about high impact giving. They publish all of their research and recommendations on their site for free, with no sign up required. They allocate your tax deductible donation to the charity or fund you choose without taking a cut. I personally give to the Against Malaria Foundation, which distributes bednets to prevent malaria, at a cost of about $5 to provide one net.
If you've never donated to Givewell's recommended charities before, you can have your donation matched up to $100 before the end of the year or as long as matching funds last. Just go to Givewell.org and Pick podcast and the Jollyswagman Podcast at checkout. Make sure they know that you heard about Givewell from the Jolly Swagman Podcast. To get your donation matched, that's Givewell.org select podcast, and then select the Jolly Swagman podcast at checkout.
You're listening to the Jolly Swagman podcast. Here's your host, Joe Walker.
Ladies and gentlemen, boys and girls, Swagman and Swaggett, welcome back to the show. It's been a minute. Apologies yet again for the radio silence. I have been a busy boy. It turns out startup life is unrelenting. Who knew? But this time I really am back for good. But enough about me. You can expect a new episode of the Jolly Swagman podcast every fortnight or so, but a flurry of activity in the next few weeks as I release some epic conversations I've recorded recently. Also expect some writing of mine to be published over the Christmas break. To make sure you get notified when it's published, sign up to my mailing list at my website thejspod.com. That's thejspod.com okay, that's enough housekeeping for now. I'll have more to say in coming episodes.
Let's just get straight into this conversation, which I recorded in Boston on the 6 December 2022.
I am here at Harvard today with Steven Pinker. Steve is a cognitive psychologist and one of the great public intellectuals of our age. He's authored many famous books, including The Blank Slate, the Better Angels of Our Nature and Enlightenment. Now, but the first book of Steve's I read was The Sense of Style a Manual for Good Writing, which exemplifies its own advice, and in which I learned that new information should come at the end of the sentence, and that one of Steve's favorite Yiddish words is Bubba Meisa. The last book of his I read is Rationality, which is going to be the focus of our conversation today. Steve. Welcome to the Jolly Swagman Podcast.
For fun, I'd like to start with writing to My Lights. You're one of the very best nonfiction writers in the English speaking world. I'm not sure if you remember this, but four years ago I asked you by email for advice on the process of writing a book, and you very kindly replied, quote, everyone is different in their writing habits. I've found my method is to write sequentially from the first to the last sentence and as intensively as possible, almost all day, seven days a week, till I'm done. End quote. Could you elaborate on why that approach works best for you? And does any of your research into the human mind support that approach, or is it just arbitrary?
Well, it couldn't support it in this if people differ, so I couldn't root it in any universal property of human cognition, if one person varies from another, I suspect it's some combination of motivation, memory and personality. In my case, the cognitive rationale is that writing a book involves coordinating many considerations. There's the line of argument, the actual content. There are the sources that I feel ought to be represented or that I must consult. There's the narrative arc of the prose itself. There's the sentence by sentence construction. There's the connection to other chapters in the book. I find that when I have all of those swarming around my head at the same time, then I can do my best to satisfy them all. If I have to put it down and come back to it, then I've got to reassemble that whole collage.
So I like the intensity of having everything in my mind for a continuous enough stretch that I can actually produce what I think is a coherent bit of writing.
Right? Yeah. That that does stand out to me about your books.
They're very internally coherent, and often that doesn't happen. On the first pass, I put my books through at least six drafts before the copy editor sees it, and then there's another couple of drafts. I shouldn't speak for everyone, but I'm not smart enough to both have a coherent line of presentation of content and to make it readable and structured. I find I have to get the ideas down first, and then I can devote more attention to style. And that's why I do it in multiple passes. Sometimes I'll rearrange chapters or paragraphs or sentences. There are people, like newspaper columnists who don't have that luxury, who tap out an article and touch it up for punctuation and press send. I'm not one of them, and I suspect most people are not. But it is the gift that leads some people to become pundits, bloggers, newspaper.
Columnists on average, how many days or weeks or months does it take you to produce a first draft?
It varies a lot by books. Some of them, like The Sense of Style, went relatively smoothly, partly because I had been thinking about the issues for a long time, and there was less factual material that had to be looked up and checked. Whereas for a book like The Better Angels of Our Nature on the History of Violence, I had to educate myself in a number of fields that I was not trained in history, international relations, and affective neuroscience that is, the brain bases of emotions. So that took a great deal more The Sense of Style, I think. I had a first draft in three months, Better Angels of Our Nature. It took a year or so for the first draft.
Let's talk about rationality. You define rationality as the use of knowledge to attain goals, and you take knowledge in turn to be justified. True Belief. A definition from Philosophy 101. My first question is observation the ultimate source of our knowledge of nature? In other words, are you an empiricist in the philosophical sense?
Not in that strict sense. Certainly in the sense that you can't do it from the armchair. That is, you can't deduce the existence of Giraffes and the Himalayas and Alpha Centauri from first principles. You have no choice but to observe the world. But on the other hand, science doesn't just consist of a long list of sensory observations, particularly since a lot of the observations that scientists make come from exotic conditions of experiments and high tech instruments. Ultimately, what we want is an explanation of the world in terms of deeper principles than what we see with our own eyeballs, but constantly verified for verisimilitude and accuracy by empirical observations. So certainly the freshman first week dichotomy between rationalist and empiricist approaches to science would be far too simplistic.
What, in your view, is the correct normative model of rationality? Enrico Petrarcha wrote a review of the book in which he argued that you implicitly rejected Gigarenza's ecological rationality as a normative model. Is that fair? And am I correct in placing you as an apologist in the great rationality debate, or do you kind of issue labels there?
I think I would issue labels. I'm not sure what the suggestion means. That as an apologist for what kind.
Of in the middle of the meleeurists and the Panglossians.
Taking that label from okay, apologists. I didn't in that context is not particularly intuitive apologies. A defender of a particular position. Yeah, well, I don't think that Gigarenzer's ecological rationality was meant as a normative model. It was meant as a descriptive or psychological model. Although Gigarenzer is a critic of the idea that certain models that were taken as normative without question, such as the use of, say, of base rates as Bayesian priors. So, for example, that the statistical prevalence of a disease in a population is equated with your aprioric credence that a patient has the disease before checking their signs and symptoms and test results. That was a background assumption in a lot of the research purporting to show that humans are not Bayesian. Viegarenzer, for example, criticizes the idea that frequency in a population is a suitable Bayesian prior.
But I don't think that he would criticize the very idea that there should be normative models that are distinct from the psychological processes that a typical human uses. So I don't see ecological rationality, which, just to be clear, has nothing to do with being green or hugging a tree, but in Gigarenzer's definition and that of John Tubi and Lita, cosmodies just refers to the way that reasoning takes place in a natural human environment. So that's what ecological rationality is, but almost by definition, it's not a normative model.
Yeah, although I guess you could say that the Gigarenza's adaptive toolkit of fast and frugal heuristics perform better in certain environments than rational choice theory.
Well, yes, but that would be as soon as you use the word better, you're invoking some standard or benchmark against which to compare ecological rationality, which means that ecological rationality itself is not that benchmark. That is, as soon as you ask the question, does ecological rationality work? Does it lead to correct or justify conclusions? Well, where do those correct or justify conclusions come from? Well, they come from a normative model, so you can't get away from normative models as soon as you ask the question, how well does ecological rationality work?
So what do you think the benchmark is?
Well, one of the reasons that I defined rationality as the use of knowledge to pursue a goal. And one of the reasons that my book Rationality is divided into chapters is that there are different normative models for different goals. Depends what you're trying to attain. So if you're trying to end up with a quantitative estimate of your degree of credence in a hypothesis, then Bayes rule is the way to go is the suitable normative model. If you're trying to derive true conclusions from true premises, then formal logic is the relevant normative model. So the normative models are always relevant to a goal, and they are also suitable to an environment in which some kinds of knowledge are or are not available.
Your graduate advisor, Roger Brown, once read a review of Lolita, arguing that it was a trove of linguistic and psychological insights. I'm curious, are there any works of fiction that provide, in your view, incredibly rich depictions of your view of human rationality?
Yeah, I'm glad you brought that up, both because Roger Brown doesn't get the attention that he deserves as a pioneer in psycholinguistics and social psychology and a gifted writer. I did write an obituary of his that's available on my website. And his review of Lolita is a masterwork published in a psychology journal, a journal of book reviews, which the editor agreed to accept as long as there were psychological implications. And Roger's review of Lolita is a true gem. So, let's see. Well, I'll start close to home. I learned a lot from at least two novels by my other half, the philosopher and novelist Rebecca Goldstein. Her first novel, The Mind Body Problem, featuring a graduate student in philosophy doing a dissertation on the mind body problem, who also faces her own mind body problem.
I'll leave it to readers to identify what that problem is, but that actually exposed me to ideas in philosophy, including, for example, the concept of mathematical realism in philosophy. The other protagonist of the novel is the husband of the graduate student in philosophy noam, Himmel, a brilliant mathematician. And in the course of their dialogues, it comes out that he, like a majority of mathematicians, it turns out, believes that mathematics is not just the manipulation of symbols by formal rules, the formalist approach, although some do adhere to that. But a majority of mathematicians believe that mathematics is about something, some abstract feature of reality that they are discovering, not inventing. That's just an example of one of the philosophical ideas that came out in the dialogue.
The other one is her novel 36 Arguments for the Existence of God, a work of fiction which is my not so unbiased view, a brilliant and very funny novel about religion and belief in God, and the title alludes to a book written by the protagonist, a psychologist of religion. His book has his best surprise bestseller had an appendix listing 36 classic arguments for the existence of God and their reputations. And the novel, a bit, I guess, postmodernly or self referentially, has as its appendix the appendix to the fictitious characters bestseller, when it came to recording the audiobook audible, had Rebecca had me narrate each of the 36 arguments, and Rebecca narrate the responses. But it's actually, I think, the best guide to the philosophy of belief in the existence of God, and it is woven into a genuinely funny and moving story.