Found in Translation: Inferencing Breakthroughs Power iFLYTEK Voice ApplicationsJanuary 24, 2018
People have dreamed of a pocket-size instant translation device at least since the days when the original Star Trek series was on the air. They’ve pined for a way to decipher the handwriting of doctors for even longer.
Thanks to advances in deep learning inferencing, iFLYTEK, one of China’s largest tech companies, has come up with practical solutions for both. In inferencing, a trained neural network is put to work to do whatever task it’s learned.
Speak to Me
The iFLYTEK portable translator is exactly what it name implies: a compact, handheld device that can translate the speech it hears into another language.
About the size of a remote control, the GPU-accelerated device translates Chinese, English and Uyghur in real time, thanks to the NVIDIA Tesla P4 and P40 training and inferencing accelerators and iFLYTEK’s advanced translation algorithm.
“It’s like having a professional translator in your pocket, helping break through language barriers, wherever and whenever you need,” said Zhijiang Zhang, vice director of the Cloud Computing Institute at iFLYTEK, China’s largest speech technology provider.
15X Efficiency Boost for Inference
By using Tesla P4 and P40 GPU accelerators, speech recognition accuracy for iFLYTEK’s Voice Cloud Platform reached 97 percent, and its inferencing efficiency rose 15x over what was possible with CPUs, Zhang said. The company expects those breakthroughs to lead to new business applications and business models.
Since its launch in March, the portable translator has sold in 97 countries and provided services more than 2 million times, iFLYTEK said.
Smart Medical Care with AI
iFLYTEK isn’t confining its efforts to translation. Working with several Chinese hospitals and medical institutions, the Hefei-based company is combining its expertise in speech technology and AI to make medicine smarter for doctors and patients.
Forget about struggling to read your doctor’s handwritten notes or prescription. Instead, doctors can simply dictate information, and iFLYTEK’s technology automatically converts the recording into electronic medical records.
The company says its intelligent dictation — deployed on the Tesla P4-accelerated iFLYTEK Voice Cloud — filters out unnecessary information to quickly transform medical data into text. Its Chinese speech-to-text accuracy is 97 percent, the company said.
Using its Voice Cloud Platform to process massive amounts of data in the cloud, iFLYTEK has launched numerous intelligent speech services, including IFlyDictation, iFLYTEK Voice Touch and Lingxi Voice Assistant. The Voice Cloud Platform has more 1.5 billion users, with a daily service volume reaching 4 billion.
“As the iFLY Voice Cloud continues to be upgraded and its scope of application expands, the NVIDIA deep learning GPU platform has given technological momentum to iFLYTEK, and has allowed users to immediately benefit from state-of-the-art technology and enjoy a better user experience,” said Zhang.
You can learn more about iFLYTEK by attending our GPU Technology Conference, March 26-29, in Silicon Valley. Xiaodong Tao, CEO of iFLYTEK Healthcare, will discuss how the company is building on its work in AI and voice recognition technologies to help doctors. Register now.