231-922-9460 | Google +

Monday, October 27, 2008

Turning Audio Into Words on the Screen

The only transcription services we've seen in action are the court reporters on TV legal dramas. So when we wanted to turn some digital audio files into a written document, we went hunting online.

It turns out there's a surfeit of transcription companies and freelancers: What was once a cottage industry that provided assistance to a handful of professionals -- think doctors taking "notes" by talking into tape recorders -- has now become a sizable business tapped by a wide range of users, from podcasters to students seeking to dissect a professor's lecture.

Fueling the demand over the past few years: the advent of digital audio (no more tapes to send back and forth) and the rise of outsourced labor (much of the work is done on the cheap by companies with offices abroad, particularly in India). Plus, in the online era, there's simply increased interest for all things written: Consider that within minutes of the conclusion of Tuesday's presidential debate, Web sites were posting transcripts of the exchange between the two candidates.

To put the new generation of transcribing firms to the test, we took the same set of interviews we recorded with chefs at a cooking competition, totaling 111 minutes of audio, and had it transcribed by five different companies we found through the Web. Most companies make it simple to submit the digital audio files online. To complicate matters, we chose interviews we knew posed problems because of background noise, speakers with thick foreign accents and a subject matter -- the culinary arts -- with plenty of semi-obscure terms and technical jargon.

We quickly discovered just how competitive the field has become. We were quoted rates ranging from a low of $55 to a high of $277.50 for the work, based on a turnaround time of one week. Many of the services can expedite jobs for a higher fee. But we eventually learned that price is not necessarily tied to quality. For that matter, we learned that regardless of cost, no service is 100% accurate.

Take the case of our most expensive firm, Michigan-based Oasis Office Support (which markets itself via the site GetItTranscribed.com). The company, which charged $277.50, ended up having a greater error rate than some others, mangling not only the occasional challenging proper noun (a restaurant by the name of "Cafe du Parc" became "Cafe Debark"), but also basic phrases ("tomato salad" became "Camilla Salad"). The company also dropped words and phrases constantly. Granted, in most cases, it indicated the missing text in the transcript with a time marking, so we could easily find the passage ourselves on the audio file. But other transcribers were able to fill in a few of those same blanks.

A couple of services in the middle of the price spectrum delivered middle-of-the-road results. ISource Solutions, which promotes itself via the site TranscriptionStar.com and employs a team of up to 60 transcribers in India, charged $129.50. It seemed determined to capture every last phrase (the transcript was filled with lots of prefaces or interjections, such as "Right" and "Well, you know"). But it didn't make much of an attempt to understand the subject matter, so that "nouvelle" cuisine became "new vel" cuisine and "allsorts" licorice became "old sorts" licorice. It also was the only service to miss its promised delivery time, coming just a few hours past its three-day mark.

With GMR Transcription, another California company, we appreciated the company's frequent updates on the progress of our job, which ran $236.50: At any point, we could log into our account on the firm's Web site and see the status of each individual audio file, from "uploaded" to "assigned" to "downloaded." Plus, we got email alerts when each file was ready to be viewed. The transcribing was spot-on in some instances (some unusual last names -- say, Asprinio -- were spelled accurately), less so in others ("County" Road became "Downey" Road). Most problematic of all: In one instance, some comments were attributed to the wrong speaker.

If there was one company that seemed to take the most pains with accuracy, it was New Mexico-based CastingWords, which specializes in podcast transcriptions but welcomes clients of all kinds. A company spokesman says they build in "a very large number of review stages" and "keep most of our transcribing in the United States." (Companies that use transcribers based in India or elsewhere abroad will concede that American English is very much its own "language.") The attention to detail showed in our order: We were constantly surprised that phrases or names that other companies missed, from "tabula rasa" to "turducken" (it's a Cajun specialty), were identified and spelled correctly. Moreover, the transcript pointed to key non-verbal details in an exchange, such as the laughter that followed a remark. But its accuracy came with a price: Our order ran $249.75, making CastingWords the second-most-expensive company we tapped.

And the cheapest? It was a service we found through Elance, an online company that connects customers to freelance contractors in a variety of fields, from graphic design to, yes, transcription, through a bid process. After posting our job, we heard from about a half-dozen services. We decided to take a chance on the cheapest provider, who charged $55.50 and offered the fastest turnaround time (a mere two days). While the work was far from perfect -- more "old sorts" licorice and the like -- the service often performed better than our most expensive one. Clearly, it pays to shop around.