New IBM Capabilities Reduce Speech Recognition Training Times from a Week to 11 Hours
One of the key challenges of Machine Learning currently is the issue with developing strong and reliable speech recognition. Conventionally, training Natural Language processing models take a very large repository of billions of words and thousands of hours of speech, along with computer hardware powerful enough to process this plethora of information in an acceptable time-frame.
IBM is trying to solve this problem by deploying a distributed architecture and boost processing power. The software giant is confident that its framework will speed up processing times by 15X without losing accuracy. The backbone of IBM’s concept is made up of several graphics cards with the company’s researchers confident of the training time drops as real.
Voice Assistants need a sophisticated encoding-decoding mechanism for them to function. They are also on the larger side which makes it very difficult for them to train at scale. IBM’s Deep Learning Method asynchronous decentralized parallel stochastic gradient descent (ADPSGD) aims to boost the volume of the repository being fed into Voice Assistant learning models. The company is using something they call as a ‘principled approach’ to ensure accuracy in the model in addition to faster learning times.
“Turning around a training job in half a day is desirable, as it enables researchers to rapidly iterate to develop new algorithms,” Zhang, Cui and Kingsbury wrote. “This also allows developers fast turnaround time to adapt existing models to their applications, especially for custom use cases when massive amounts of speech are needed to achieve the high levels of accuracy needed for robustness and usability.”
With companies as prominent as IBM in the software industry map pitching in to enhance progressive AI capabilities such as Machine Learning and Deep Learning, products developed in these technologies will get a fresh lease of life. Currently, industry behemoths such as Google, Amazon, Microsoft all have their Voice Assistant capabilities. Amazon has gone a couple of steps further by launching Alexa for business along with home, and, now also for Healthcare. However, user sentiment is not very positive about the product’s reliability. Maybe IBM’s new initiative will make these machines efficient, propelling an entire AI-powered industry.