As firms in the US gear up for Commodity Futures Trading Commission’s (CFTC’s) communications recording rule 1.35(a), Anthony understands the headaches involved in converting voice recordings to text.
For Christmas, my parents got me a pretty amazing gift—Dragon NaturallySpeaking.
Dragon is a speech recognition tool. Where Apple’s Siri understands and responds to speech, Dragon listens to and then transcribes it.
I love being a journalist and hope to spend my career in this profession, but the one thing I absolutely abhor is transcribing interviews. Some people have the gift of being able to listen and type accurately, all in real time. I'm not like that. I record every interview I conduct because I can't listen, formulate questions, understand what's being said, and type all at the same time.
At best, it takes me about 90 minutes to transcribe a 30 minute interview. It is a nightmare. And if there's an accent involved, it can take twice as long. I'm so bad at transcription that I have been known to pay others to do it for me.
So with Dragon, I had hoped to become a more productive journalist by cutting down on the time it takes to transcribe interviews. Sadly, that did not happen.
While Dragon does a good job of interpreting my words correctly, it is abysmal at understanding the interviewee's voice. It's not completely useless, though: I can dictate the interview into the device, which might save me a few minutes in the long run.
So, transcription is sadly still a headache, albeit a necessary one that I accept as part of being a reporter.
Voice Recording for Compliance
Compliance officers reading this might be nodding their heads in sympathy. In the upcoming February issue of Waters, my colleague Jake Thomases looks at Commodity Futures Trading Commission (CFTC) Rule 1.35(a), which mandates that firms trading commodity interests keep records of all communications pertaining to those transactions for up to one year. This includes email, instant messages and—crucially—audio.
The feature, titled "Wired for Sound,” will be available online next week. It's an expansive look at new enhancements in the expensive and challenging art of capturing, storing and analyzing voice recordings. One section, though, jumped out for me:
The best way to do any analysis of audio is to transcribe it to text first. But as any reporter can tell you, the transcription software on the market today is less than perfect. Banks are familiar with it, having employed it in their call centers for years, whereupon they would flag certain stock names that came up in conversation. But call center operators are trained to speak slowly, articulate well, and wait until the other person is finished speaking before they start. That makes it easy on the transcription software. Conversations on the trading desk aren't nearly as civilized. One way to approach the problem is to record in stereo, producing an A leg and a B leg for the two parties that can be transcribed separately. But even that falters when there are three or more parties on a call.
If you don't absolutely have to accurately transcribe conversations, then it's just not worth the time and—in the case of capital markets technology—the expense. In Jake's feature discusses some of the doubts industry participants have as to how closely the CFTC will audit this rule.
If there's not enough of a prod—i.e., if the penalties are not more severe than the headache and expense of installing the technology—then firms will likely continue to play the wait-and-see game when it comes to 1.35(a).
On a side note, though, if any of you vendors working in this space have a beta version of a transcription device you would like to test out on a journalist, please email me at [email protected] or give me a call at 646-490-3973. I'll be your guinea pig.
Jesse Lund talks about real uses for DLT in the capital markets, lessons learned while rolling out IBM's blockchain platform, and what’s ahead for 2018, and into 2019.Subscribe to Weekly Wrap emails