How Twine Enhanced Audio Analysis of HyperSentience's AI model

February 29, 2024

The Overview

HyperSentience is a pioneering AI company based in London, specializing in context awareness technology. By processing audio from smartphone devices, they empower phones, VR/AR headsets, smartwatches, speakers, and laptops to comprehend users' day-to-day activities and adapt seamlessly. To further improve their technology, HyperSentience sought natural audio recordings to train their AI effectively.

"Working with Twine has been an exceptional experience. Their ability to consistently deliver data that drives our project forward is commendable. The level of service, professionalism, and dedication to understanding our needs set them apart. We're looking forward to more successful collaborations in the future."
– Ian Sherwin, Product Owner at HyperSentience 

Problem Statement

HyperSentience requires a diverse set of natural audio recordings capturing individuals engaged in various day-to-day activities within their homes. The challenge lay in ensuring that participants performed these activities genuinely, without staging or planning, to facilitate authentic data collection for AI training.


To address HyperSentience's requirements, we devised a comprehensive solution encompassing various factors crucial to capturing genuine audio recordings. The key components of our approach included:

Phone Distance: Differentiating the distance between the mobile phone used for recording and the activity being performed, categorized as < 1m, < 3m, < 6m, or > 6m, allowing for varied perspectives during activities.

Phone Mounting: Identifying how the phone was mounted during the recording—whether flat on a surface or held in hand—for an accurate analysis of audio quality.

Location: Noting the room where the activity took place, the specific location within the room, and whether the recording phone was in the same room.

Background Content and Volume: Capturing types of background noise present during the recording, such as traffic noise, music, or room ambiance, along with measuring the general volume of background noise before recording.

Description: Participants provided comprehensive step-by-step notes detailing the occurrences during the recording to facilitate efficient annotation.


To ensure optimal audio recording quality, we guided participants through setting up the correct audio settings on their smartphones. The phones were placed in natural positions, as they would be during daily activities. The microphone's sensitivity allowed for placement at a reasonable distance from the event, eliminating the need for proximity to loud sounds.


Twine played a pivotal role in generating audio recordings that comprehensively encompassed all the elements required to train AI on improving audio analysis of daily activities. Beyond effective project management and freelancer recruitment to create this dataset, Twine employed its secure vault payment system to efficiently handle all freelancer payments.

As we continue to drive AI innovation through ethical data collection and labeling, Twine remains committed to delivering exceptional service and fostering lasting partnerships with visionary companies like HyperSentience. Together, we are at the forefront of advancing AI capabilities for context-aware technology.

Related Customer Stories