aiOla launches Jargonic, the most accurate speech recognition model for both academic benchmarks and real-world enterprise use

Outperforming competitors like OpenAI and ElevenLabs across standard datasets and specialized jargon recognition, Jargonic sets new standard for conversational AI

aiOla, a leading conversational AI lab, today announced the launch of Jargonic, its enterprise-grade speech recognition model that outperforms all competitors across both academic benchmarks and real-world business environments. In comprehensive testing, Jargonic achieved the highest accuracy on standard datasets and outperformed competitors in the areas most critical to enterprises adopting AI speech solutions, establishing it as the industry’s most accurate speech-to-text solution available. Developed by aiOla’s world-class research team led by renowned speech AI experts Prof. Yossi Keshet, Dr. Gil Hetz, and Prof. Bhiksha Raj, Jargonic is now available via API through aiOla’s enterprise platform, enabling organizations to integrate best-in-class speech AI capabilities into their operations, processes, and applications.

While OpenAI’s Whisper and proprietary speech recognition models post impressive benchmark scores in controlled environments with pristine audio samples, real-world enterprise settings present drastically different challenges. Enterprise environments frequently involve low-quality audio, significant background noise, diverse accents and dialects, and specialized industry jargon rarely found in training datasets. This gap means that systems that excel in laboratory conditions often fail when deployed in factories, warehouses, field operations, or customer service environments.

aiOla’s Jargonic model directly addresses these real-world challenges by being specifically designed for enterprise environments. Trained on over 1,000,000 hours of transcribed speech – including substantial data from industrial settings – Jargonic incorporates proprietary keyword-spotting technology that seamlessly detects and accurately transcribes domain-specific terminology.

In comprehensive benchmark testing, Jargonic achieved the highest average accuracy across four gold-standard English academic datasets with an average word error rate of just 5.91%, outperforming leading competitors including Eleven Labs (6.14%), Assembly AI (6.25%), OpenAI’s Whisper (6.52%), and Deepgram Nova-3 (6.48%).

Beyond academic benchmarks, Jargonic excelled when tested on real-world financial earnings calls featuring specialized terminology from diverse industries. When focusing specifically on industry jargon terms, Jargonic correctly recognized nearly 9 out of 10 specialized financial terms (89.3% recall), significantly outperforming other leading models including Eleven Labs (85.1%), Assembly AI (82.9%), Whisper (80.1%), and Deepgram (77.5%). For overall accuracy on jargon-heavy financial speech, Jargonic achieved the lowest error rate at 12.32%, compared to competitors ranging from 13.09% to 17.73%.

Jargonic also demonstrated exceptional multilingual jargon recognition, achieving over 95% accuracy in recognizing industry-specific terms in the CommonVoice dataset across German, Spanish, Portuguese, French, and English—consistently outperforming all competitors in every language tested.

“The real challenge in enterprise speech recognition isn’t achieving high scores on clean academic datasets—it’s maintaining exceptional accuracy when confronted with specialized vocabularies and challenging acoustic conditions,” said Prof. Yossi Keshet, Chief Scientist at aiOla. “Our keyword spotting technology represents a fundamental breakthrough in how speech recognition models identify and process domain-specific terminology.”

“Enterprises today are sitting on vast amounts of untapped spoken data that contain critical business intelligence,” said Assaf Asbag, Chief Technology and Product Officer at aiOla. “With Jargonic, we’re not just offering incrementally better speech recognition – we’re fundamentally redefining how organizations capture, understand, and leverage their spoken communications.”

aiOla’s research team is led by Dr. Gil Hetz, VP of AI, and Prof. Yossi Keshet, Chief Scientist—both of whom bring extensive industry experience in speech synthesis, including notable work on the Alexa research team. The team is supported by Prof. Bhiksha Raj, aiOla’s Distinguished Scientist, who additionally leads the Voice Lab at Carnegie Mellon University. Together, this group is driving global innovation at the forefront of speech AI.

Credit: aiOla