Google has submitted an article to the scientific community about its advances in speech synthesis. Thanks to its Tacotron 2 system, it would have managed to generate an audio stream with a result so natural that it could not be distinguished from that produced by a human voice.
Tacotron 2 is based on two neural networks. The first divides the text into sequences, and transforms each one into a spectrogram. That is, a diagram visually representing the intensity of the audio frequencies. The second, named WaveNet, automatically generates audio files. Its particularity is that it was originally developed by DeepMind, the famous artificial intelligence company owned by Alphabet. WaveNet already makes Google Assistant's English and Japanese voice work by using an unsupervised learning method: it does not rely on a large database fed by recordings of actors. This is a slightly modified architecture version that has been used here. WaveNet acts as a vocoder to synthesize the temporal waveforms of the spectograms.
Tacotron 2 is so powerful that it can handle difficult-to-pronounce words without problems and is able to subtly vary the volume such as speech velocity. It also modifies intonation as a function of punctuation, accentuating, for example, terms written in capital letters.
A little regret all the same, Tacotron 2 is expressed for the moment with a female voice and in English. To speak with another feminine voice or a male voice, it would be necessary that the artificial intelligence is driven again. It still has a little way to go before benefiting applications like Google Assistant.
The first surgery assisted by Hololens is French
This is a world first, and as often when it comes to health, French people are in the spotlight. A surgery operation helmet HoloLens (the laying of a shoulder prosthesis on a patient of 80 years) was held on December 5, 2017 at the Avicenne Hospital, located in Seine-Saint-Denis. Dr. Thomas Grégory was able to scan the area to operate in 3D ... even before the operation, thanks to the use of Microsoft's HoloLens helmet.
On December 5, 2017, Microsoft's Hololens augmented reality glasses were used for the first time in orthopedic surgery. An octogenarian patient was placed a shoulder prosthesis by Dr. Thomas Gregory, Avicenne Hospital Bobigny in the Paris region. The Hololens eyewear allowed medical teams to virtually overlay a 3D digital model of the patient. The goal ? "See what the eyes can not see," says Gregory.
This digital model projected on the patient is a concentrate of information collected during imaging examinations: radiography, scanner or MRI. "This allows us to have a lot of information during the operation: the exact thickness of the tissue, the precise location of invisible surrounding organs, relevant 3D constructions or 2D slices.The tool allows us to increase the senses of the surgeon and allow to be faster and more accurate, "says the surgeon. So many details that allow a more careful intervention and a better recovery of the patient. For Dr. Gregory, this world premiere must "show how HoloLens glasses revolutionize surgery for the benefit of patients."
With regard to orthopedic surgery, the ability to make the skeleton appear through the skin is invaluable. Or to see such a small invisible vessel, and that it is not necessary to section. But all areas of surgery are potentially concerned. Pr Stephan Haulon, surgeon of the aorta at Marie-Lannelongue Hospital (Hauts-de-Seine) hopes he can perform a vascular intervention (on the blood vessels) within a few months. "Augmented reality puts us at the dawn of a real revolution, because we will be able to integrate a number of information that we could not integrate until now." This is not virtual surgery, but the increased surgery, and that's really what's changing our practices, "he insists.
In addition to having access to these data on the operated patient, Prof. Haulon sees HoloLens as having a pedagogical interest for both patients and students. "During the consultation that precedes the surgery, I can show the images to the patient, also equipped with glasses: where is his aorta, how will the intervention be performed, etc. In the same way, in terms of training, the device allows to explain to the students the pathology and how to treat it from images in 3D or even in 4D that we share together at the same time.There is also a 'gamification' effect in the training which is very motivating, "adds Dr. Thomas Gregory for whom,Simulation is the future of learning surgery.
Part of the world premiere of December 5 was broadcast live. Dr. Gregory finally adds that the device allowed him to be supported at a distance by three other surgeons: an American, a British and a South Korean. Each with a particular area of expertise. "They see what I see and what I realize during the intervention, and can give me visual indications or enlighten me in case of questioning," he explains.
There remains a crucial point to develop in the logic pursued by these pioneering surgeons and Microsoft: the registration of the digital model depending on the position of the patient or possible movements, even very small, during surgery (breathing, tissue deformation) the passage of surgical tools ...). Indeed, the projected image is based on examinations made at a time T. "This is the next step, on which we are already working: dynamic registration," explains Stéphan Haulon. He whose specialty is to navigate the blood vessels to pose stents endoscopically precise: "We sail fairly rigid launchers that will distort the vessels and dynamic registration, in real time, would correct or s to adapt to these deformations.There are several ways to do this: to redo 3D acquisitions on a regular basis, but this implies a high radiation rate for the patient, or to use ultrasound imaging, which allows some markers to reconstruct an image. We also do a lot of computer modeling to predict the deformation according to this or that gesture.It remains impossible to say today which technique will be used tomorrow to reach this goal.But we are aware of this limit. is a few millimeters, but we are sometimes close to the millimeter. " But more than the extra precision, it is the time of the intervention that could be improved. "The digital model is finally used as GPS in 3D.It is a valuable navigation tool that should allow us to save a lot of time," says Stéphan Haulon. A crucial parameter when it comes to unblocking an artery following a heart attack or stroke.
No comments:
Post a Comment