AssemblyAI Releases Conformer-1 API Speech Recognition Model

AssemblyAI has launched Conformer-1, a speech recognition model trained on 650,000 hours of audio data.

This model attains an accuracy level comparable to humans, reducing errors by up to 43 percen when processing noisy data, and speeding inference time by 29 percent and training time by 36 percent, according to the company.

In developing this model, the company also leveraged Efficient Conformer, a modification on its original architecture.