People of all races around the world, regardless of age or gender, do voice labeling. If a new collection of voice files is needed, it can be done at Uptempo. 

 

We collect from the voices of young children to middle-aged, elderly, and even animal sounds in a variety of fields: Radio, Voice Over, Animal Sounds, Dubbing, Youtube Video, Nature sound, and music.

 

 

Quality control for building voice data

Uptempo Data team adheres to the ESSENTIAL 5 principles that contain our own know-how. And based on this, we build image data that guarantee optimal quality.

 

Step 1: Raw data collection

  • Collecting voice data in a file format such as MP4 (including SMI text data)
  • Filtering and removing silent sections and unnecessary data that are not suitable for processing and utilization purposes

Step 2: Source data building

  • Data labelings such as detailed data classification and de-identification of collected raw data
  • Standardization and setup with source data in a form that can be processed through crowd working

Step 3: Source data processing

  • Primary processing: Marking of the target section within the target work voice data
  • Secondary processing: Securing voice data within the displayed area and building text (Label)

Step 4: Processed data inspection

  • Complete inspection: Implementing basic quality inspection across the construction data by applying strict quality standards
  • Cross-checking: Implementation of grouping and cross-quality inspection by applying the K-fold cross-validation method

Step 5: Final delivery of construction data

  • Client final delivery only for final data of suitable quality
  • For the insufficient amount of construction, only high-quality data is delivered by constructing more than 150% of excess data