IOLA, speech recognition innovation... automatically selects the optimal ASR model based on the context

Artificial intelligence startup aiOla has launched a groundbreaking new solution that can revolutionize speech recognition accuracy. The company’s released “Speech Intelligent Gateway” system can analyze user speech in real-time and automatically connect to the most suitable speech recognition model. The system dynamically assesses complex language features to select the model that achieves the best accuracy for processing.

Last year, aiOla introduced “DRAX,” a speech AI model that overcomes traditional speech recognition limitations through parallel stream learning technology. DRAX can process all statements simultaneously and performs strongly in environments with background noise, intonation variations, and other real-world variables. Building on this technology, the newly released “QUASAR” analyzes speech features, speaker intonation, noise presence, context, and other information to automatically select the most appropriate model from numerous automatic speech recognition engines.

Although several ASR service providers such as OpenAI’s Whisper, Amazon Transcribe, Alibaba’s Qwen2, and Deepgram are competing in the market by optimizing for noise environments or intonation, most companies still rely on a single model that performs best in standard evaluations. This results in frequent recognition errors in real-world applications, leading to ongoing criticism over user experience.

aiOla co-founder and President Amir Haramaty pointed out the current situation where companies are forced to accept the limitations of specific ASR models: “Some models excel at handling American English but often struggle with British accents or noisy environments.” He emphasized, “QUASAR is the first system to treat speech recognition as a dynamic problem rather than a static technology.”

In internal benchmarks, aiOla applied this system to various real-world scenarios involving different accents, background noises, and specialized content. Results showed that in 88.8% of response requests, the system could dynamically select the optimal ASR engine to improve accuracy. It is expected that this technology will significantly enhance understanding in fields such as customer support, meeting transcription, and automated response systems.

Haramaty stated, “As speech recognition increasingly becomes the fundamental interface connecting humans and AI, recognition errors have become unacceptable.” He called QUASAR a “technology that transforms ASR into a living infrastructure,” adding, “This is not just a technological breakthrough but a transformative shift that can impact everything from global call centers processing billions of calls to independent developers creating subtitle functions.”

aiOla plans to leverage this technology to greatly improve the practicality and reliability of speech AI interfaces, creating a structural turning point for the entire AI speech ecosystem.