Breaking: aiOla Surpasses OpenAI's Whisper


Breaking News: VentureBeat Reports aiOla Surpasses OpenAI's Whisper in Jargon Recognition!


aiOla Outperforms Leading Speech Technologies with 95% Accuracy

For businesses seeking the use of speech recognition tools to help complete workflows hands-free and seamlessly, accuracy is the most critical requirement. Although there is an array of technologies to choose from in the market, most solutions are either designed for personal application or lack the required understanding to effectively function in business settings. 

That is, until now – aiOla’s AI-powered technology has achieved something that no other speech technology has yet to do. 

aiOla is the first-of-its-kind to possess the ability to understand business-specific jargon in over 100 languages, any accent, and any acoustic environment with near perfect accuracy. 

In fact, aiOla outperforms OpenAI’s industry-leading Whisper model by 7%. Based on a benchmark of keyword and jargon detection using a dataset of 16 world languages, Whisper’s largest model yields 88% accuracy. aiOla rises above 95% accuracy and does not require any training to reach this level of accuracy! Against another benchmark of hard-to-detect keywords taken from English language audiobooks, the CED model from Apple’s research yields 92.7% accuracy, whereas aiOla reaches 95.1% accuracy. 

So, what sets aiOla apart? 

Put simply, its proprietary and patented keyword spotting model, AdaKWS (a moniker for its Adaptive Instance Normalization Keyword Spotting model). AdaKWS is groundbreaking because it is able to pick up on words that it has never before trained on, even when they are highly specific to a business’ unique workflows.

Whereas OpenAI’s Whisper is trained on small snippets of audio, aiOla’s AdaKWS is trained upon complete sentences, which more closely resembles how people actually converse. For example, when put into practice, this is the difference between a speech recognition tool knowing what “DVIR” (or, driver vehicle inspection report) means during process completion within the fleet industry, or skipping over the imperative acronym entirely. If the latter occurs, the workflow may be inaccurately completed or perhaps stopped in its tracks, rendering the use of a speech-recognition tool utterly useless. 

Without getting too technical, aiOla’s model applies a character-based encoder with parameters that tune into an adaptive module to detect the keyword. By doing so, aiOla can pick up on niche vocabulary and decipher it from regular conversation. In just a few hours, any company can be onboarded and apply aiOla to its mission-critical workflows. 

aiOla automatically generates micro language models for use, so workers can complete their operations using nothing more than speech. In turn, business processes are completed with greater efficiency and safety, as workers complete tasks hands-free. Additionally, aiOla captures otherwise lost data, which is really important to have available for deep insights, pattern recognition, and even collaboration across office locations, both locally and globally. 

“The greatest challenge for keyword spotting at this time is the lack of training data that’s available and relevant for business-specific use cases. Typically, businesses have to fine-tune existing models to train the technology to be able to recognize jargon. aiOla surpasses existing solutions with accuracy levels of over 95%, so any business can recognize its quantifiable ROI upon its first use,” says aiOla’s Chief Scientist Professor Joseph Keshet.

Here’s an example in Japanese:

When aiOla is used to help complete any existing or new business process through speech, the model will detect and understand business-specific jargon without any issue. This is crucial as over 50% of vocabulary used in business settings is unique to either the company, process, or industry. As a result, any company can implement aiOla and remain confident that the solution will perform exactly as it is expected. By being able to adapt to any industry’s unique jargon and custom workflows, aiOla has been transformative across industries, including: automotive, defense, aviation, finance, insurance, healthcare, manufacturing, logistics, and more. 

“While we have already overcome some of the greatest hurdles that face the speech-recognition and AI industry, we are only just getting started. This ability to transform real-time speech into correct and safe actions, as well as actionable insights, has the power to augment every industry from the ground-up,” expresses Amir Haramaty, aiOla’s CEO and Co-founder. 

Learn more about aiOla and how you too can turn speech into action and ROI or read more about this research on VentureBeat.