Starting state: building a dataset and manually annotating it
Lalilo is an online phonics program, providing an interactive tool for teachers to assist with customised reading learning for young kids. They are building a speech recognition system to detect mispronunciations for their read aloud features. Because very few corpuses of annotated recordings of children exist, Lalilo had to build their own dataset. They then started annotating the data internally. They noticed rather quickly that their turnaround time for the large volumes of data they needed to annotate was not sustainable so they started looking for a partner to help them with this.
Accelerating data annotation with super.AI
They reached out to us to help accelerate their data annotation. Thanks to our Humans & AI solution, were were able to handle a much larger volume of data and annotate it at a significantly faster turnaround time. Thanks to our large crowd, we were able to provide Lalilo with a higher number of annotators than they had internally and produce a higher quality of accuracy for the annotated data. Lastly, we were able to provide a customised solution to them, adjusted to their data inputs and providing the data output in their desired format.
Thanks to our collaboration, Lalilo was able to significantly scale their internal dataset of recordings while improving the required accuracy needed to train their speech recognition system.