Project Overview

The product team aimed to introduce a new feature that identifies and highlights potential deception within agent-client conversations. This feature would analyze spoken dialogue and provide deception hints alongside the transcription. I was tasked with redesigning two existing wireframes to enhance the usability and learnability of this functionality, prioritizing improvements to the transcription screen.

Enhancing User Experience in Voice Analytics: Deception Detection

Screenshot of an audio transcription interface showing a conversation between an agent and a caller about a traffic accident, including a highlighted statement 'I did not see the cyclist using the crosswalk' in red.

Client: Voice Analytics Startup

Industry: Voice Analytics

Project Duration: 10 Days

My Role: Market Research, Prototype Design, Wireframe Redesign

Company Size: 11-50 employees

Challenges

  • Tight Timeline: Only 10 days to deliver a high-quality wireframe redesign.

  • No Direct User Access: As a short-term contractor working on a new feature, I had no access to user data or direct user interviews.

  • Complex Feature: The transcription screen, which required significant attention, integrates audio playback with deception flagging, demanding intuitive design.

Screenshot of a web interface showing a list of files with details such as call ID, score, status, file name, upload date, duration, location, department, case number, and an assign button.
Screenshot of a call transcript with highlighted text, a file attribute panel on the left showing details like file name, call notes, date/time, duration, size, call type, location, department, and risk level. The call transcript contains spoken dialogue and timestamps, with some words highlighted in red.

Table View

Audio and Transcriptions (Highest priority)

User Research & Ideation

To compensate for the lack of direct user data, I conducted informal interviews with users of similar products to uncover critical workflows:

  • Analysts frequently toggle between audio and transcription, often at increased playback speed (~1.25x).

  • They focus their analysis on flagged segments indicating potential deception.

  • Taking notes directly on cases is common to support decision-making.

Visual benchmarking of audio/video streaming platforms revealed best practices for player placement and navigation flow, influencing my design approach to surface the audio controls prominently and streamline transcription interaction.

Design Iterations

Iteration 1

  • Moved audio playback controls to the top of the screen to align with natural eye movement and promote ease of use.

  • Introduced separate audio tracks for agent and claimant calls, recognizing the dual source nature of conversations.

Iteration 2

  • Shifted from displaying full transcripts to highlighting only deception-flagged segments, reducing cognitive load.

  • Added timestamp and speaker identifiers for clarity.

Computer screen showing a file management system with a list of digital files, including file names, statuses, upload times, sizes, and locations with links to download.
Screenshot of an audio transcription interface showing a conversation. The conversation involves a caller describing a car accident where they did not see a cyclist using a crosswalk, and the agent asking about noticing the cyclist beforehand. The interface displays audio playback controls, timestamps, and options to follow up or close out, with navigation icons on the left side and labels for language and help.
Audio transcript of a conversation about a car accident shows the caller explaining that they did not see the cyclist using the crosswalk, and the agent asking if they noticed him before the accident. The transcript includes timestamps and highlights the words "I did not see the cyclist using the crosswalk" and "No,". 
Table view
Audio & transcriptions
Screenshot of an audio transcription interface showing two audio files labeled Claimfile9876.wav and Agent. The Claimfile audio has red and green segments, while the Agent audio is in blue. There are options for playback speed and rewind 15 seconds. The interface includes menu icons for home, files, link, language, help, and profile on the left, and buttons for follow-up and close out on the top right. The transcriptions include placeholder text, Latin lorem ipsum, with timestamps. 
Mid-fidelity Prototype - Iteration one
Screenshot of an audio transcription interface with two audio files and chat transcripts. The first file is named 'Claimfile9876.wav' playing for 2.22 seconds of 9.32 seconds. The second is named 'Agent' playing for 1.22 seconds. The chat includes text: 'Lorem ipsum dolor sit amet?' and 'Lorem ipsum.' 
Mid-fidelity Prototype - Iteration two

Final Prototype

  • Introduced a commenting feature enabling analysts to annotate cases inline for future reference.

  • Refined layout for clear hierarchy between audio controls, transcripts, and notes.

Audio & transcriptions - Comment feature

Impact

  • Enhanced feature usability by aligning design to analyst workflows, improving task efficiency and focus.

  • Reduced information overload through targeted transcript display, accelerating deception detection.

  • Improved stakeholder confidence with a polished, data-informed design ready for implementation.

Reflection

This project highlighted the vital role of user research, even informal channels, in guiding AI product design under strict constraints. It reinforced the value of clear articulation of design decisions when navigating stakeholder collaboration. Additionally, it deepened my expertise in designing interfaces that balance complex data with streamlined user experiences in AI-powered tools.

Thank you for reading!