Baidu Deep Voice real-time conversion from text to Voice; GTX 1080 TI was released with super Titan X performance | AI developer headline-Alibaba Cloud Developer Community

content preview:

automobile and baidu Deep Voice to realize real-time conversion from text to speech

according to Lei, Baidu announced Deep Voice today, a product-level text-to-speech (TTS) system.

The system is completely composed deep Neural Network built, the biggest advantage is that it can meet real-Time the conversion requirements. In the past, the speed of audio synthesis was often very slow, and it took several minutes to hours to convert several seconds of content. But now, Baidu Research Institute has been able to realize real-time synthesis, on the same CPU and GPU, the system is 400 times faster than the original audio waveform depth generation model DeepMind released by Google WaveNet in last September.

However, at present, Deep Voice need the help of a phoneme model and audio synthesis components. The Baidu R & D team hopes to achieve end-to-end speech synthesis in the future.



**Facebook supports pre-trained word vectors in 90 languages.

Do you still remember FastText? This is the open-source tool Facebook released for large database text processing. Today, the FastText research team released their latest research results on GitHub-using Wikipedia for training, including 300-dimensional word vectors in 90 languages; All using FastText default parameters for training.

The FastText team expressed the hope that all developers can provide feedback. In addition, a large number of new models will be released soon, please pay attention to Lei Feng website (Public number: Lei Fengwang) follow-up reports.

Supported language list and download address:

automobile and avida releases GTX 1080 TI, with performance exceeding Titan X

today (the evening of the 28th local time), Avida announced at the GDC conference GTX 1080 TI. The graphics card still adopts Pascal 16 nm process technology, with thermal design power consumption of 250W and 11GB GDDR5X memory. Avida said its performance exceeded GTX 1080 by about 35%. This makes GTX 1080 TI and Pasal Titan X have the same performance, and some non-public models may even have the same performance.

In addition, the public appearance of GTX 1080 TI is the same as that of GTX 1080. However, Avida announced at the press conference that the heat dissipation of its public version has been improved, including removing the DVI interface to increase the air outlet area of the graphics card; Under the same noise level, compared with the heat dissipation of the public version on 1080/1070, the effect of the new solution is improved by about 5%.

GTX 1080 Ti will be listed next week at a price of only US $699 (about RMB 4808). The first batch is only available in public. GTX 1080 sold in the United States has begun to reduce its price. According to Lei Feng's website, many retailers have reduced their prices to $100.

Although GTX 1080 TI mainly focuses on the game market, but for deep learning developers, they can buy almost the same computing performance at half the price of Titan, which is very tempting. Currently Nation tour price has not been announced, Lei Feng snare to continue to pay attention.


daily Recommended reading

master practice drills, top 10 machine learning time series prediction problems

this is the latest practical practice for developers organized by our old acquaintance, Australian machine learning expert Jason Brownlee. These are ten challenging time series prediction questions. For these ten difficult problems, the classical linear statistical method is not enough to solve, and the high-order machine learning method must be used.

These challenges come from Kaggle.

If you want to challenge your difficult exercises, improve your ability to process time series datasets, and practice machine learning technology-these ten questions are for you.


Python code implementation of basic machine learning models and algorithms from scratch

Erik Linder-Norén, a foreign Machine Learning Developer, uploaded the Python code he used to create various machine learning models to GitHub for sharing. Let's take a look at other people's models. Of course, these algorithms really start from scratch, are very basic, and have the greatest reference value for beginners.

The model code uploaded by Erik Linder-Norén includes decision tree, logistic regression, multi-layer perceptron, random forest, and support vector machine.


related articles:

Google Summer of Code list of 2017 partners; Intel 12-core Skylake-X | AI developer headlines

Selected, One-Stop Store for Enterprise Applications
Support various scenarios to meet companies' needs at different stages of development

Start Building Today with a Free Trial to 50+ Products

Learn and experience the power of Alibaba Cloud.

Sign Up Now