DataoceanAI helped a leading technology company deliver approximately 2,000 hours of natural duplex conversation data in five months, coordinating more than 1,000 participants across recruitment, recording, annotation, quality assurance, multi-channel synchronization, and compliance.
Key Results
The Challenge
A leading technology company needed a large-scale natural conversational speech dataset to support the continued development and optimization of its voice interaction technologies.
Unlike scripted or read speech, natural conversations capture how people actually communicate, including changes in speaking pace, contextual responses, conversational feedback, and natural speaking habits.
The project required approximately 2,000 hours of data to be produced within five months, involving more than 1,000 participants. The scope covered participant recruitment, recording, data processing, annotation, quality assurance, and final delivery, while also meeting strict participant authorization and GDPR-related compliance requirements.
The challenge went beyond scale.
Participants had to meet specific criteria related to age, gender, occupation, and language capabilities. Meanwhile, data collection, annotation, and quality assurance needed to progress in parallel despite changing workloads.
Multi-microphone recordings added another layer of complexity, as audio captured across different channels had to remain accurately synchronized for subsequent processing and calibration.
The Solution
Scalable Participant Recruitment
DataoceanAI recruited participants through online channels, offline resources, and partner organizations, screening candidates against project-specific demographic and language requirements.
This multi-channel approach enabled the team to organize more than 1,000 qualified participants while maintaining the flexibility needed to respond to changing resource demands.
Automated verification tools were also used before and after collection to check participant uniqueness and identify duplicate voice samples, helping reduce the risk of repeated or invalid data.
Flexible Production and Quality Control
Instead of relying on a fixed production line, DataoceanAI dynamically allocated resources according to actual workloads.
Data collection, processing, annotation, and quality assurance were carried out in parallel. When recording volumes increased, additional annotation capacity could be introduced; when quality assurance workloads rose, more QA specialists could be deployed.
During peak periods, 300 to 500 annotators or quality assurance specialists could be mobilized to help prevent bottlenecks and keep production aligned with the overall delivery schedule.
Quality assurance was conducted throughout production rather than only at the end, enabling issues related to audio quality, data completeness, annotation, and task execution to be identified and addressed early.
Multi-Channel Synchronization and Compliance
For multi-microphone recording tasks, DataoceanAI used professional acoustic laboratories or controlled recording environments to maintain stable collection conditions.
Hardware-level clock synchronization kept audio channels aligned within the millisecond range, providing a reliable foundation for downstream data fusion, calibration, and processing.
Participant authorization, identity information management, and data processing requirements were also incorporated into the workflow from the outset to support compliance with project-specific authorization standards and applicable GDPR-related requirements.
The Results
Within five months, DataoceanAI completed the production and delivery of approximately 2,000 hours of natural duplex conversation data involving more than 1,000 qualified participants.
By coordinating recruitment, recording, annotation, quality assurance, synchronization, and compliance within one integrated production workflow, the project maintained stable production despite changing workloads and a compressed delivery schedule.
For the customer, this reduced the operational complexity of managing a large participant network and coordinating multiple production stages.
The delivered dataset met the required participant profiles, quality requirements, and authorization standards, providing a reliable data foundation for subsequent training, evaluation, and optimization of voice interaction technologies.
Build Conversational AI on Data That Reflects Real Communication
DataoceanAI provides end-to-end data engineering support for complex conversational AI projects, from participant recruitment and natural speech collection to annotation, quality assurance, synchronization, and compliant delivery.
Talk to DataoceanAI about your next conversational AI data project.
