Combining speech-to-text AI and human effort to create bespoke transcriptions

Kensho Scribe's Human-In-The-Loop service combines AI speed with human editing to deliver transcripts that are accurate, perfectly formatted, and ready to use.

When most people think of audio transcription, they think of a simple AI-handled task. But how would you transcribe the following audio?

Here are a few variations:

“As of 12/31/2022, our company has generated ½ million dollars in profit”

“As of Dec. 31st 2022, our company has generated $500,000 in profit.”

“As of December 31 2022, our company has generated half a million dollars in profit”

The reality is different automatic speech recognition ASR vendors will give you different results on an audio clip like this.

With Kensho’s Scribe’s Human-In-The-Loop (HITL) service, you receive an AI-assisted human-edited transcript where every word is transcribed the way you intend it to be and every page looks the way you intend it to look.

Scribe HITL provides many customizations, including:

Visual styling

Sometimes there’s information in your transcript you want to include that goes beyond the words spoken. The Human-In-The-Loop service gives you the flexibility to add your company logo, custom speaker stamps, timestamps, coloring options, and more. A transcript you receive from the HITL service is custom tailored so that you can send it directly to an end consumer of the content.

Language nuances and style guides

Between variations between British and American English, omissions of small talk, omissions of filler words, conversions of contractions, Oxford commas, there are various language specifications you may want to specify for your transcript. When using the Scribe HITL service, you will work with a dedicated Kensho account team to ensure all your language needs are known and transcribed correctly.

Omissions

Organizations that deal with calls that containing highly sensitive information, compliance concerns, or even superfluous chatter often want to customize their transcriptions to find and omit certain portions, such as identifying information, non-public matters, and irrelevant chatter. This work is made possible when you add trained humans to the process. Kensho Scribe, through its relationship with S&P Global, employs a full-time staff of human reviewers, who go through four months of transcription and security training, to securely and efficiently flag and omit the portions of transcripts a client would not want written.

Numbering

Numbers can be tricky. Scribe will not only transcribe the numbers the way you prefer, but also can contextualize and fact-check figures. Take a variation of the earlier example:

“Our company did five hundred last year”

The Human-in-The-Loop team can research externally to contextualize this “five hundred.” If it was expressed in dollars, yen, pesos, and if it was expressed in thousands, millions, or billions.

Sometimes AI isn’t enough. When it comes to styling, there is no one-size-fits-all solution.

At Kensho we hold the belief that AI isn’t here to replace humans, but empower them. If you’re looking for high quality, styled, 99.9% accurate formatted transcripts, Scribe HITL may be the best solution for you. Talk to us about your unique transcription needs.