In the arms race for speech-enabled technologies, systems that can support the widest user base will win out in the end. That’s why when a Fortune 500 Tech company sought comprehensive speech training data in French, they needed to ensure that the collected data would truly account for the wide range of dialects present in the francophone world.
That meant DefinedCrowd needed to source 600 hours of speech from 1,000 unique speakers who represented a cross-section of genders, ages and regional dialects.
To obtain 600 hours of speech, we created an extensive library of unique prompts covering the client’s specified domains. To do so, we tapped into our French-speaking Neevo community, who produced written examples of how they might express intents (such as disputing a charge or returning a package) in 15 words or less. We collected and validated those variants and used them as a basis for the web-crawled text that filled out our prompt library, leaving us with more than 200,000 prompts in the end.
We gave those 200,000+ prompts to our Neevo contributors, who read and recorded them via Horatio, a mobile app developed in-house and available across Android, iOS and Windows devices. We sourced a representative cross-section of 1,000 French-speakers who read and recorded the prompts. Another group of 2,000 contributors validated every second of the 600 audio-hours the first group had produced.
Our services enabled our client to bring a better solution to a competitive marketplace faster and at a far lower cost than would have been possible if data had been handled in-house. At DefinedCrowd, we aren’t satisfied by simply playing the role of service provider. Our philosophy is to work alongside our clients to serve as true data partners. For this project, that meant we were right next to the client every step of the way, first helping them to define their needs, and then setting up the workflows, gathering and structuring collected speech data, and ensuring the quality of the results through expertly-made qualification tests and automated analysis. It’s this philosophy that allows us to deliver training data with: