In the realm of voice solutions, spontaneous speech performance is crucial for Automatic Speech Recognition (ASR) technology to operate effectively. In this article, we will discuss how our client, a leading speech recognition technology company, leveraged Defined.ai’s transparent metadata access to improve their ASR models’ spontaneous speech performance.

The customer

Our customer, a leading speech recognition technology company, develops Automatic Speech Recognition (ASR) models for their customers to implement voice solutions.

The context

The client reached out to Defined.ai seeking large amounts of speech data from various locales to train their ASR models and, more specifically, to improve their ASR models’ performance on spontaneous speech received via a telephone line. Due to the time-sensitive nature of the project, the AI training data had to include high-quality transcriptions with accurate timestamps and speaker metadata, all for immediate consumption. The client highlighted a need for ethical data sourcing and expressed interest in learning how data collection and data protection aspects were handled.

The solution

Defined.ai’s off-the-shelf offerings had everything that the client required. To ensure the data was exactly what the client needed—nothing more and nothing less—we shared the full metadata of the relevant datasets at the recording, speaker, and turn levels. Using this information, the client was able to precisely select recordings demographically representative of the distributions of age, gender, and accent essential to their modeling needs. Defined.ai then customized a price and terms based on their specifications before packaging and securely delivering their custom data selection within three business days. The client gained much-needed trust and confidence because the entire process, from start to finish, was transparent and straightforward.

To ensure the data was exactly what the client needed—nothing more and nothing less—we shared the full metadata of the relevant datasets

By leveraging Defined.ai’s high-quality datasets, which are compliant with international data privacy protections and wage policies, the client not only positioned themselves as a responsible leader in the industry but also delivered exceptional results to their customers.

Defined.ai’s support enabled the client to make strategic decisions, leading to tangible improvements such as a significant reduction in the word error rate across various locales supported by their ASR solutions. By effectively utilizing the high-quality speech training data we provided, the client achieved these outstanding results without encountering the typical delays in model deployment caused by resource-intensive training data preparation. This efficiency enabled the client to adapt and thrive in a competitive market while delivering top-notch performance.

Ready to optimize your ASR technology’s spontaneous speech performance? Discover how Defined.ai’s metadata access and tailored data collection services can help you achieve exceptional results. Check our Marketplace today to learn more!