aiOla, an Israeli startup, has made significant strides in addressing the limitations of traditional speech recognition models by introducing a method that enables these systems to understand industry-specific jargon and vocabulary. This innovation is particularly important for complex enterprise settings where typical speech recognition models often falter due to challenging acoustic environments and specialized terminology. aiOla’s approach involves a two-step “contextual biasing” technique that first identifies domain-specific keywords using its AdaKWS model and then integrates these keywords into the ASR decoder to enhance overall accuracy.
In initial tests, aiOla adapted OpenAI’s Whisper model, resulting in a marked improvement in word error rates and overall detection accuracy. This approach can be applied to any speech recognition model, including Meta’s MMS and proprietary systems, making it versatile across various industries. The startup’s adaptive model has already been deployed with Fortune 500 companies, showing significant time and cost savings in processes that involve technical jargon. aiOla’s research is open for further development by other AI teams, although the company currently offers access to its technology only through a subscription-based product suite.
