UlatusCase StudiesScaling Multilingual Audio Data Collection for AI Model Training
Scaling Multilingual Audio Data Collection for AI Model Training

Scaling Multilingual Audio Data Collection for AI Model Training

How Ulatus Delivered 2,400+ High-Quality Audio Files Across 26 Locales as a Strategic AI Data Partner

Audio Data Collection Project
Technology / Artificial Intelligence

Project Summary

Service:Audio Data Collection for AI Model Training
Language Pairs:Multiple source languages, 26 locales
Volume:2,400+ Audio Files, 650+ Hours
Domains:Healthcare, Finance, Call Centers
Timeline:2 Months
Industry:Technology / Artificial Intelligence

About the Client

Our client is the global AI audio solutions team at one of the world's leading technology and mobility giants. They partnered with Ulatus to develop large-scale multilingual audio datasets to train and improve their speech recognition and AI models across 26 locales.

The engagement came with a strict 2-month deadline. High quality recordings, precise technical specifications, and consistent daily delivery were all non-negotiable.

Challenges we faced during the project

Natural Sounding Conversational Speech

The recordings needed to sound like natural conversation rather than scripted reading, since the audio was being used to train speech recognition and AI models. This had to be achieved consistently across all 26 locales and across the Healthcare, Finance, and Call Center domains.

400+ Resource Hiring in 2 Months

The project required hiring more than 400 contributors within the two-month timeline. Because the audio had to cover 26 locales, the recruitment had to reach the right speakers for each one, including the harder to source accents.

650+ Hours of Delivery

The client required more than 650 hours of finished audio. This volume had to be delivered on a consistent daily schedule so the client's AI model training timeline stayed on track.

Per-Person Audio Minutes Restriction

Each contributor could only record a limited number of audio minutes. Reaching the total volume therefore had to be spread across a large number of contributors rather than a few.

Recording Tool Limitation to Meet the Scale of Requirement

The recording tools available were not able to meet the scale the project required. A different recording approach was needed to reach the volume within the timeline.

Solutions provided by Ulatus

01

Software to Record Multiple Conversational Meetings at One Time

A software setup was used to record several conversations at the same time rather than one at a time. This increased the amount of audio that could be captured within the project timeline.

02

100+ Hours of Delivery per Day

The project maintained a delivery pace of more than 100 hours of audio per day. This pace was sustained throughout the project, including the holiday period.

03

Benefit from Global Office Network

A global office network was used to source contributors across regions. This helped cover all 26 locales, including the harder to source accents.

04

Quick-Response Tech Team to Automate Custom Project Management Workflow

A quick-response technical team built automation into a custom project management workflow. This was used to track production, manage the delivery pipeline, and support quality checks.

Accomplishments

2,400+ audio files delivered, expanded from an initial scope of 1,300 files.

650+ hours of high-quality, AI-ready audio data produced across all 26 locales.

Coverage achieved across 3 domains: Healthcare, Finance, and Call Centers.

Consistent daily delivery maintained throughout the project, including the holiday period.

Client AI model training timelines were met without disruption.

Recognized by the client as a strategic data collection partner.

Improved audio data quality and volume compared to previous vendors.