Head of Transcripts Wade Baker Talks Speech APIs, the Cadence Community, and the Future of the Transcription Industry


Greg Eiselt

When you think of transcription, you probably think of courtrooms where court reporters create transcriptions of proceedings in real time, or cloud-based machine-learning software that churns out the written version of an audio file in less than 10 minutes.

However, it's a little more complicated than that. For many Cadence clients, transcriptions of calls are more than just written outputs of audio; they're detailed research documents that may make or break a multi-million (or multi-billion!) dollar investment decision.

That's why in February, Cadence made the decision to hire Wade Baker as Head of Transcripts, an appointment we believe will greatly help us reach even more clients seeking world-class, human-powered transcription. We caught up with Wade to learn about his transcription background and how he sees the state of the industry.

How did you get into transcription?

I've always been a word nerd, grammar geek, and fan of languages. I'm actually a trained classical musician and spent my entire educational career studying voice, opera, and music. While I was in grad school in New York on a path to become a professional classical musician (specifically an opera singer!), I needed a job. I answered an ad for an overnight transcription company that did production work for the television industry. After I left New York, I got a job as one of the first remote employees for a major transcription company. Up until then, it was all bike messengers and cassette dubs; there was no such thing as "file delivery." This was in the year 2000, so remote files were sent via AOL and dial-up!

Over the next 20 years, I built my own client base, which has included vendor relationship managers for Fortune 500 companies, CEOs, politicians, and more. At this point I'm personally able to maintain 150 WPM typing speeds, which adds up to roughly 300 pages of transcript in a day. If you map that out over the last 25 years that I've been transcribing, I've personally transcribed over 1 million pages in my career.

What are some of the most common misconceptions you hear about transcription?

I think the most common misconception is that it's a low skill activity. After 25 years, I can definitely tell you it is not! It's very much a high-skill reason and language activity; the means through which we produce that language just happens to be our fingers.

The skill that those unfamiliar with the industry most often associate with transcription is typing speed, which is almost irrelevant to me as I'm looking to hire new transcribers to join our team. What's most important to me is to have a keen ear for the nuance of language. If you're fast or you've used court reporting shorthand to program your computer, then that just increases your take-home pay. You give yourself a raise by getting faster, but you don't improve your brain by getting faster.

A lot of transcription firms feature machine-powered transcripts. What is the state of machine-generated transcripts today and how do you think humans will be able to continue to add value in the future?

I think machine transcription has its place. As I mentioned, it depends on what you need the end product to be. If you are okay with a relatively inaccurate product with a high error rate, then the machine transcription is justified. For products that need to be used by humans and read by an end user, however, machine translation alone is simply not able to cut it. While the machine can recognize raw phonemes, I'd argue there's much more to language than just recognizing a word being spoken. It looks like word soup unless a human has operated over it.

For example, if a human stutters and says a word three times, a machine isn't smart enough to know if they're being emphatic and emphasizing something by saying it three times or if they're stuttering and have just repeated themselves. The machine can only represent exactly what was spoken; there's no reason there. The machines also struggle with punctuation. If you've ever read a direct machine transcription, it's quite difficult to read.

Again, this might work fine for low-touch items which don't have business impact. However, if the transcript is informing something with business impact, the error rate in machine transcription is just too high right now for people that have real need for accuracy or for people whose documents need to be read by an end user. At the very least they still need to be heavily edited, and I think it actually takes longer for a skilled editor to edit a machine transcription versus a skilled linguist trained in transcription to produce the final document.

Tell me about Cadence. What made you decide to join the team as Head of Transcripts?

Well, I've been laboring in my little ivory word tower alone for over 20 years now with no colleagues. While the flexibility of this setup is great, the one thing that's missing when you're operating like that is human connection. When I found Cadence and realized that there was actually this robust community with colleagues that I could ask questions of and people active on social channels (while completing work), it felt like an office environment that gave me a social component that I hadn't had before. The collegial atmosphere of cooperation is also quite unique to the industry; it's not just about meeting raw deadlines.

I also really love how challenging the material is. What I like about Cadence is that there's a new topic to learn about every day. Whether it's date farming in Saudi Arabia, semiconductor manufacturing in China, or US auto retailer CRMs, there's always a new and interesting challenge that requires research, which helps keep me from becoming a mindless automaton.

Finally, as Head of Transcripts, I get the opportunity to leverage my experience to build a full-fledged department from the ground up, a (slightly daunting!) undertaking that I've loved so far.

What do you think Cadence is doing differently from other transcription providers?

I would say our greatest differentiator is that we hire linguists who are language experts and can operate with reason over language. Our first priority is not how fast can you type. Our first priority is, are you a language expert who is skilled in research? Yes, we want transcripts out the door fast, but what's essential for us to create is a research document that our end client will find useful. Indeed, Cadence's goal is to facilitate a high level conversation. Whether we facilitate that conversation via our foreign language speaking interpreters and translators or through our English-speaking transcribers who inform a decision with a research document, our job is to facilitate communication for people who are making real business decisions.

Can you give us a sneak peek at some of the features planned for the future?

We're always listening to what our end clients are looking for to help their investment decision-making process. One feature we're rolling out now for select clients is a client portal that allows all interactions, including requests, processing, and delivery of transcripts, to be done through a secure platform instead of being exchanged via email.

A few other requests from clients we've received that are in the works include:

  • Audio embedded within the transcript that allows the reader to listen to any part of the transcript audio with one click;

  • Link indexing at the beginning of the transcript via a table of contents

  • Inserting associated visual assets, e.g. Powerpoint slides, in corresponding sections of the transcript;

  • Compliance flagging for quicker identification of material, non-public information.

Ultimately, our clients value our flexibility and dedication to their success more than any specific feature or add-on. This flexibility and dedication are what allow us to provide an amazing product for our clients and why I'm so excited to see what I can help build as part of this wonderful team.