“OK Google”. “Hey Siri”. We are all speaking to our smartphones and listening intently to what they have to say. Although their synthesised voices are beginning to sound less and less robotic, they still have a long way to go before they sound completely human-like.
Just this week, Baidu announced their new technology called Deep Voice, which seems to be a great leap forward in making these things sound like humans. Do you think synthesised speech that sounds completely human-like is going a step too far, approaching the uncanny valley? One thing which is not discussed in the Baidu Research is the gender of the voice: do you think it’s time you had more choice in male versus female voices? Tweet us your thoughts @cognition_x.
At JPMorgan Chase & Co., a learning machine is parsing financial deals that once kept legal teams busy for thousands of hours.
The program, called COIN, for Contract Intelligence, does the mind-numbing job of interpreting commercial-loan agreements that, until the project went online in June, consumed 360,000 hours of work each year by lawyers and loan officers. The software reviews documents in seconds, is less error-prone and never asks for vacation.
Made possible by investments in machine learning and a new private cloud network, COIN is just the start for the biggest U.S. bank. The firm recently set up technology hubs for teams specializing in big data, robotics and cloud infrastructure to find new sources of revenue, while reducing expenses and risks.
In this paper, Baidu presents their latest innovation in synthesised speech. Therein, they present Deep Voice, a production-quality text-to-speech system constructed entirely from deep neural networks. Deep Voice lays the groundwork for truly end-to-end neural speech synthesis.
The system comprises five major building blocks: a segmentation model for locating phoneme boundaries, a grapheme-to-phoneme conversion model, a phoneme duration prediction model, a fundamental frequency prediction model, and an audio synthesis model. For the segmentation model, they propose a novel way of performing phoneme boundary detection with deep neural networks using connectionist temporal classification (CTC) loss.
How does consciousness arise? Researchers suspect that the answer to this question lies in the connections between neurons. Unfortunately, however, little is known about the wiring of the brain. This is due also to a problem of time: tracking down connections in collected data would require man-hours amounting to many lifetimes, as no computer has been able to identify the neural cell contacts reliably enough up to now. Scientists from the Max Planck Institute of Neurobiology in Martinsried plan to change this with the help of AI. They have trained several artificial neural networks and thereby enabled the vastly accelerated reconstruction of neural circuits.
They have developed and improved staining and microscopy methods which can be used to transform brain tissue samples into high-resolution, three-dimensional electron microscope images. Their latest microscope, which is being used by the Department as a prototype, scans the surface of a sample with 91 electron beams in parallel before exposing the next sample level. Compared to the previous model, this increases the data acquisition rate by a factor of over 50. As a result an entire mouse brain could be mapped in just a few years rather than decades.
Telefónica has launched an AI layer to its network that will allow its customers to interact with the company via a digital assistant called Aura. The move is the latest attempt by the telecoms industry to add services such as advertising and content on top of their network to compete with technology businesses including Google and Facebook but have failed to make a mark leading to accusations they are no more than “dumb pipes”.
They hope that adopting artificial intelligence technology to give customers control over the data they generate using a smartphone will open a door back into the services market. Aura will also allow the owner of O2 to cut its spending on call centres as it pushes more customers to use Aura for basic customer service inquiries.
C3 IoT updated its Internet of Things platform with more machine learning and artificial intelligence technology, integrated more with Amazon Web Services and has more than 100 million sensors and devices under management.
Version 7 of C3 IoT’s platform will help the company expand into new verticals. As noted previously, C3 IoT initially gained traction with utilities and then moved into new verticals. Houman Behzadi, chief product officer at C3 IoT, said the company is actively working on implementations at oil and gas, health care, and financial services companies.
Yesterday, Predii, the industry leader powering predictive maintenance and repairs through AI, hosted a panel to discuss The Future of AI in Transport in Silicon Valley.
The panel brought together industry leaders, including Predii executives and development partners, along with tech media representatives, research firms, and educational institutions to discuss and debate the role of AI in automotive connectivity and predictive maintenance.
Alan O’Herlihy, CEO and founder of Everseen, was able to ask scientists and researchers he has collaborated with—and who are working on AI solutions that exceed what any of us has seen so far—which industries are likely to be transformed by AI first and what that will look like.
Based on their answers, here are five industries he thinks will lead the charge: 1) medicine, 2) emergency (search and rescue), 3) public transportation, 4) education, and 5) retail. In the article, he discusses how AI will fuel their disruption and how we are currently progressing.
In this segment of Julien Despois‘s series on deep learning (check out part 1 here), he show us how autoencoders can help us visualize data in some very cool ways. He shows how this works on images, using the Convolutional Autoencoder architecture (CAE).
In the post, he presents several techniques to visualize the learned features embedded in the latent space of an autoencoder neural network. These visualizations help understand what the network is learning. From there, we can exploit the latent space for clustering, compression, and many other applications.