Voice Recognition Media

One-on-one with Trint's war correspondent CEO - as transcribed by Trint


By John McCarthy, Opinion Editor

July 24, 2018 | 14 min read

London-based audio transcription service Trint claims to be taking the heavy lifting out of interviews and making life easier for journalists.

Libya Arab Spring

Jeff Kofman reporting in 2011 during Arab Spring

Founded in 2014, the service has married audio and video transcription with AI technology and a text editor, a service journalists have been avidly awaiting for many years. But does it deliver – and does it deliver enough for 17p per minute of transcription?

Making bold claims is company chief executive and co-founder Jeff Kofman, an Emmy-winning journalist. His exploits include covering the Arab Spring, the invasion of Iraq, the Chilean miners and the Libyan Revolution. Such was Kofman's confidence that his service could even handle tricky journalistic situations in warzones, The Drum decided to issued a challenge to Trint. We would interview its chief executive if we could publish the transcripts as Trint transcribed them, to let readers review its usefulness for themselves.

Trint said yes.

Before we get onto the Q&A, some house rules.

No copy has been corrected or amended. It is the verbatim Trint transcript.

In the interest of brevity, some of the more boring or meandering bits have been slashed. And some paragraphs have been inserted to break the text. We're sure you've got better things to do than read through a 30-minute call.

The AP claims it hits accuracy levels of 95%. This wouldn't have been far off that. From there, users are invited to listen back and tidy any outliers.

The Drum: You are an Emmy-winning journalist, what was it that spurred you to found a voice transcription company?

JK: You know if you had said to me ten years ago when I was a reporter for ABC News and 15 years ago when I was in Iraq covering in Baghdad covering the Iraq war you know you're going to be an entrepreneur and inventor and run a tech company I would have said Not a chance. You don't know me. I'm a reporter through and through. I'll die in the newsroom that carry me out feet first.

I mean no I am totally astonished to find myself doing this. But you know one of the fun things about life is that if you're open to it you can surprise yourself in great ways. I just stumbled into this to be honest. I had taken a buyout from ABC News I was a broadcast journalist for more than 30 years I started in Toronto was with the Canadian Broadcasting Corporation for more than a decade.

I wanted to do something new and if I'm honest I didn't know what it was. Started working in a bookstore taught university and just by circumstantial just by by coincidence met some developers who had done some really interesting experiments with transcription and I said Oh God I hate transcribing. I've spent thousands of hours of my life doing it. It's part of my job. I like least. And know wouldn't it be amazing to leverage the increasingly accurate output of automated speech to text to see if it could do the dirty work for you.

What was it like launching Trint?

So we now have an iPhone app just to give you the history of the company. I left ABC News November 30th 2014. We started working on Trint December 1st 2014. The next day we were a team. It was me and three developers.

We had our first prototype out for testing and about three or four months spent a lot. Through my my extensive contacts in journalism and in the US the UK and Canada was able to put together a really really strong group of testers and broadcast in print and radio and media production and they were really enthusiastic to see this work so they gave us really. They gave us very strong feedback that helped shape the product and we ended up we launched the product 20 months later in September 2016 into commercial revenue.

And it's just been it's just been growing ever since. I think we were about seven or eight people then in May of 2017 just over a year ago we were 10 people. We did a sizeable funding round. We're now a team of 36 and we're hiring 11 more. We're opening a North America American headquarters in Toronto and the product now includes Trint enterprise as the Associated Press was our first big enterprise client. They're using us now globally and in multiple languages. We also launched this winter or spring a mobile app on iPhone.

It's really interesting when you're in the early stages of a startup you're selling a promise you're selling a dream and people have to believe in you. As a founder and in your team and when the product is still in prototype in development it hasn't had proven revenue when it hasn't had proven product market fit. People are looking for ways to see whether they should trust you whether they should believe in you particularly investors. And you know obviously my own track record and credibility gave people comfort that I understood what the solution had to look like because I've lived the workflow problem.

The Google Digital News Initiative, BBC Worldwide Labs and Cisco really helped Trint get off the ground, how vital was this funding?

Having the support of the BBC through BBC Worldwide labs having the support of Cisco UCL and getting investment from the Knight Foundation which is the big media foundation in the US early on all of that really it. It lets people know that people actually think that there's something there there's a there there that there's something something there. Plus winning the global editors network which is the biggest journalism conference in Europe. We won their start up your news competition in 2016. All those things incrementally give people attention.

We just found out that we want another big grant from Google to support our next one of our next phases of development and how we won our strategy. We won a large grant from Google's digital news initiative to support a translation project that we have developed and they are Google is giving us money to fund the build of it.

It's really exciting because it allows us to hire a bunch of people to do work on this project which we've already done a proof of concept on and we've shown this to a number of our customers.

So how does Trint help journalists and publishers?

Obviously there is pressure on publishers journalists and particularly for ad revenue and getting the content faster and faster. Vealers trying to that sort of a process. So you know the Associated Press tried. There are a number of tools that that claim to do what we do. We have a couple of competitors that have followed us in the market and the Associated Press looked at them and selected us after a lengthy study and we've gone through this with a number of news organizations and had come out on top.

And they did a workflow study and they found that on average we were saving. There are journalists and editors in our day [that should read an hour a day] of transcription time. So when you multiply that by 10 50 100 600 people. That's a huge amount time that's being won back by software because you know you take an example.

What surprising ways are publishers using Trint?

The Associated Press uses us in Washington. They they record State Department news conferences but don't staff them.

So that news conference can go on for an hour 90 minutes. They have the recording but it's not searchable. Then they discovered that something was said about North Korea and they need to get that sound bite the quote on North Korea. Traditionally what they did was they said to a producer an editor go take that recording listen and find the reference to North Korea.

Two hours later the producer emerges from a screening room with that quote. Now they put it through Trint. They search the word North Korea they find the quote and it takes them a minute. That's that's a very tangible example of how we've totally transformed their workflow in a totally set up working.

Trint desktop

What's your pitch to publisher's, accuracy makes or breaks a reporter's reputation?

The challenge that I set for the development team when we began was automated speech is really good but it makes mistakes. I'm a reporter I can't live with flawed data with mistakes. A lawyer a medical professional marketing person a person who corporate communications nobody can take that kind of mangled but close to accurate output of something like Siri and actually put in his or her reputation on it because we all know that you know a bus goes by a dog barks and it's going to be messy. Somebody with a heavy foreign accent or somebody interrupting the automated speech engine will make a mistake.

So you know what we're about is applying A.I. so that you can actually leverage it [the transcriptions] and it's useful and trustworthy because as accurate as A.I. is and you know with an accent like Donald Trump's or Barack Obama's to be nonpartisan or Teresa Mayes or mine I'll come out 98 99 percent accurate. But those one or two words that are wrong are our death to a journalist or anybody any other professional. If you don't capture them before you publish them.

So the challenge with A.I. is to understand that it can do a lot but don't assume that it's an it can do it all. And so let me do the heavy lifting which is what we do. And then through the Trinta editor which we developed and patented you can then polish it to perfect. So you know that's that's what we tell people. We're not claiming that a trend in its first form the transcription that you get from this conversation is going to be perfect. What we are going to say is that I'll say the word Mississippi. And if you then want to find that word in this conversation I've only said it once. All you need to do is try to search through that word and you will find it instantly and you can find that reference. [Immediately found Mississippi].

That's where it's incredibly useful and then you can say you know if I use an exotic name or a term that's not in our dictionary you can correct it. So you know that for newsrooms the the cell is you know we save you time. You know we make your employees more efficient and we get your content out faster to multiple platforms.

The point is that that in a world where more than 80 percent of the traffic on the Internet is video and audio recording which isn't searchable. We need to make it discoverable searchable accessible.

Are just journalists using Trint?

So we find no higher education has been a huge adopter for qualitative research for things like that but also for archived corporate communication and as a learning tool that's where we're starting to explore and for accessibility for students and academics who have hearing impairments or learning disabilities.

We're also seeing a big take up in media monitoring people who need to find out what's been said online and see if there are any references there that relate to their topic for example the BBC uses us to monitor Russian and Spanish language newscasts.

What is next for Trint?

We've gone through incredibly rigorous testing and that's a fundamental and that's expensive. It just is to make sure that you can you can withstand hacking and that you know that your vulnerabilities are caught before somebody else penetrates them. It takes resources. You have to have a team dedicated to it. Nobody in the media and government and law and health care will talk to you if you can't get to that point.

And you know if you're using public API for transcription for from some of the big companies they're scraping the data and that means your data is not secure. So you got to know you know not all automated transcription services are alike.


What's been your biggest difficulty in getting the company off the ground?

You know the challenge when you're a first time entrepreneur and you have no business training you kind of have to have a sense of humor because you know when I first began this and first started talking to investors I didn't know the difference between a KPI in an hour or a lie.

And you know I mean I have no business background I I know I have the vision for the product that I have the drive to get it to see it build but the business side has been a huge challenge. You know the first time I ever tried to do a cash flow. A friend built an Excel spreadsheet and left it with me and I tried to change some numbers and I kept getting hashtag hashtag hashtag and I I just got so frustrated I just wanted to curl up under the table and scream send me back to Baghdad.

Any startup founder will tell you the number of decisions you have to make every day. And even I think even if you have an MBA nothing prepares you for the endless number of strategic complex decisions you have to make. And you know one of the things that I've learned is that indecision is worse than a bad decision.

And this is something that as a reporter I had to deal with everyday. When you're in a hazardous region in a war zone or even when you're not you know you're on deadline you've got to decide do I go left do I go right. And sometimes you don't know. But sometimes it's okay let's go left and you discover well that's not the village I thought it was going to be.

How wary should publishers be when embracing tools to increase efficiency and workload?

It becomes really difficult to sort out what tools will actually effortlessly fit within our current workflow and create savings of time and money in efficiencies and which ones are going to promise to do that but create as many problems as they claim to solve.

And so what the Associated Press said was that they are very wary of what they call a slice of bread a sink an innovation or an app that does just one thing. Because if it does that it's got to talk to a lot of other technologies and it's in it and it's up it's left to them to figure out how to do that. They said one of the reasons we went with Trint is because you guys were already doing a whole bunch of things.

You're a loaf of bread and we watch it grow bigger and bigger each month. So you know I think the caution for for news organizations and for journalists is make sure that this miracle solution that's being presented to you actually doesn't create as many problems as it solves.

Voice Recognition Media

Content created with:

More from Voice Recognition

View all


Industry insights

View all
Add your own content +