Beyond Coding: Transcription Excellence at Parrot, a Glimpse into Our Workflow

Historically, Depositions were recorded by court stenographers. While their transcriptions generally boasted an accuracy of about 95%, they still needed manual revision or third-party finalization. Our AI, while not perfect, can identify the current speaker or even detect when an interpreter is translating a non-English response. Unlike traditional methods that take seven to ten days, we can provide a revised, notarized version within 24 hours, often at a reduced cost. It's worth noting that stenography is a dwindling profession in the US, with approximately 10% of the current 23,000 specialists retiring annually.

A Dual Approach for Different Needs

Our challenge was creating a platform that catered to both our clients who needed annotation and collaboration tools without modifying the actual content. and transcribers, focused on revising and editing. Legal professionals often prefer printed testimonies, referencing specific lines in documents during court sessions or discussions.

Different target groups mean different workflows. For our transcribers, efficiency is paramount. The more they can review and correct in a shorter span, the better for our business growth and their earnings. We've designed an editor that's attuned to their working style. As they listen to the audio recording and simultaneously review the AI-generated text, they correct any discrepancies. This combined audio-video-text structure presents both technical challenges and interesting design considerations: Click on a word in the text, and the audio starts playing from that point. As the audio plays, each word gets highlighted in the text, mimicking a 'karaoke' effect. Search functionality? For transcribers, this is only a simple front-end CTRL+F in-memory search issue, as they are still editing the text. For clients, the final text doesn't change, making it a back-end matter. This distinction allows us to integrate third-party search engines, incorporating advanced features like typo tolerance and semantic searching.

A New Era in the Judicial System

While we encourage our clients to adopt a digital approach over the outdated method of physical copies, there are situations where they might need tangible data. For instance, when they lack internet access in court or can't use their computers. We allow clients to create "clips" from the integrated audio, and sometimes video, with the text. They can compile these clips into one track on our platform and download it, be it audio or video with embedded subtitles from our transcript. This makes the evidence more credible in court due to its full format and minimizes the chances of distorting the meaning.

In the future, we aim to improve our 'Auto Summary' feature, which, using AI, can distill hours-long testimonies into a concise few-page overview. Similarly, we take responsibility for our outputs and approach product quality with utmost care. Changing long-standing habits and convincing people to trust technology often meets skepticism. Thus, delivering maximum quality, addressing bugs promptly, and swiftly implementing improvements are critical for us.

The judiciary deals with vast amounts of data, often processed manually. Through machine learning and AI, there's immense potential to speed up, simplify, and enhance these processes. For example, analytics could offer insights into the case progression, effective questioning techniques, or process efficiencies. Broadly, it's about processing enough data and training the machine learning model to interpret it. The outcome could be considerably faster legal disputes, bringing a new era to the global judicial system.

Closing Thoughts

At Parrot, our mission is to revolutionize litigation, marrying human expertise with cutting-edge technology. If you're a software engineer passionate about transforming industries and pushing boundaries, consider joining our dynamic team. Together, we can continue pioneering innovations in the transcription realm.