Custom voice, a specific text to speech (TTS) service in commercial speech platforms, aims to adapt a source TTS model to synthesize personal voice for a target speaker using few speech data. Custom voice presents two unique challenges for TTS adaptation: 1) to support diverse customers, the adaptation model needs to handle diverse acoustic conditions that could be very different from source speech data, and 2) to support a large number of customers, the adaptation parameters need to be small enough for each target speaker to reduce memory usage while maintaining high voice quality. In this work, we propose AdaSpeech, an adaptive TTS system for high-quality and efficient customization of new voices. We design several techniques in AdaSpeech to address the two challenges in custom voice: 1) To handle different acoustic conditions, we use two acoustic encoders to extract an utterance-level vector and a sequence of phoneme-level vectors from the target speech during training; in inference, we extract the utterance-level vector from a reference speech and use an acoustic predictor to predict the phoneme-level vectors. 2) To better trade off the adaptation parameters and voice quality, we introduce conditional layer normalization in the mel-spectrogram decoder of AdaSpeech, and fine-tune this part in addition to speaker embedding for adaptation. We pre-train the source TTS model on LibriTTS datasets and fine-tune it on VCTK and LJSpeech datasets (with different acoustic conditions from LibriTTS) with few adaptation data, e.g., 20 sentences, about 1 minute speech. Experiment results show that AdaSpeech achieves much better adaptation quality than baseline methods, with only about 5K specific parameters for each speaker, which demonstrates its effectiveness for custom voice. Audio samples are available at https://speechresearch.github.io/adaspeech/.

If you have trouble reading this email, see it on a web browser. It has been a little while since we sent our last newsletter. In this edition, we are bringing you some exciting goodies we think you will love. To get started, this research paper on Liquid Time-constant Networks led by Ramin Hasani et al. from MIT showcases novel recurrent neural network models that can change their underlying equations to adapt to new data inputs to reduce complexity massively continuously. Have you tried out expert.ai's natural language API demo (no signup needed to try it!).

The U.S. Postal Service will finally get new high-tech mail delivery trucks. The agency said Tuesday that it awarded a 10-year multi-billion dollar contract to Wisconsin-based Oshkosh Defense to replace it's aging fleet of vehicles. The new fleet will replace and expand the existing more than 230,000 vehicles – among them approximately 190,000 delivery trucks – including many that have been in service for 30 years. The deal calls for the postal service to order between 50,000 to 165,000 new delivery trucks featuring 360-degree cameras, advanced braking, with front- and rear-collision avoidance system that includes visual, audio warning and automatic braking. What car was Tiger Woods driving in accident?:Golf

Microsoft is opening up limited access to a text-to-speech AI called Custom Neural Voice, which allows developers to create custom synthetic voices. The tech is part of an Azure AI service called Speech. Companies can use the tech for things like voice-powered smart assistants and devices, chatbots, online learning and reading audiobooks or news. They'll have to apply for access and gain approval from Microsoft before they can harness Custom Neural Voice. The tech can deliver more natural-sounding voices than many other text-to-speech services, according to Microsoft.

The Internet Archive announced that it has made Computerworld's print issues from 1967-2014 available online in better quality than ever. The publication provides a historical record of sorts for computing, as it launched during a key period in the industry and documented significant events like a 75 minute Steve Jobs interview in 1995. Much of Computerworld was already available on the Internet Archive, thanks to cooperation from publisher IDG and its Patrick J. McGovern Foundation. However, those were digitized from poor-quality microfilms that "did not look very good," according to the Internet Archive's blog. Luckily, some better-quality microfilm collections were acquired and donated to be digitized and preserved on film. The Internet Archive worked with open source communities and NextScan to improve the digitization by "adjusting for faded film, straightening the pages, performing optical character recognition" and more.

The first generation RPA focuses mainly on structured data, where data extraction is straightforward and it usually results in just 30%-40% Straight Through Processing (STP). In an effort to bring structure to the unstructured data, enterprises turn to cognitive automation technologies, which is a convergence of RPA and AI and ML. RPA use cases are content-dependent and in cognitive automation models, the idea is to make the RPA bots learn from human behavior. For this, vision technology like optical character recognition (OCR), document extraction tools, ML or a combination of these capabilities are leveraged to bring structure to the unstructured and semi-structured data. The major challenge in automating data extraction is due to the presence of voluminous unstructured dark data.

Many of the most successful and widely used machine learning models are trained with the help of thousands of low-paid gig workers. Millions of people around the world earn money on platforms like Amazon Mechanical Turk, which allow companies and researchers to outsource small tasks to online crowdworkers. According to one estimate, more than a million people in the US alone earn money each month by doing work on these platforms. Around 250,000 of them earn at least three quarters of their income this way. But despite many working for some of the richest AI labs in the world, they are paid below minimum wage and given no opportunities to develop their skills.

The advent of technology has brought convenience to life. Believe it or not, survival without technology is one of the darkest thoughts that can cross your mind in the digital era. The world has become a global village thanks to rapid digitization, but it has also opened doors for many fraudsters to step in and terrify people. Organizations in every sector are unsafe due to increasing ransomware and data breaches. Considering the increasing number of frauds, companies opt for robust verification systems with OCR technology to only onboard legitimate customers.

A revision to the postal law was enacted Friday, allowing Japan Post Co. to scrap Saturday and next-day deliveries of ordinary mail as early as autumn next year. The bill to amend the postal law was approved at a plenary session of the House of Councilors. The move reflects a decline in mail volumes and is also aimed at improving post office personnel's work conditions. The government had delayed the submission of the bill to the Diet in order to allow the Japan Post Holdings Co. group to prioritize dealing with the issue of irregularities in sales of Kampo postal life insurance products at post offices. The revised law will reduce the frequency of general mail deliveries from at least six days a week to at least five days a week.

The Ministry of Information and Communications (MIC) launched the akaBot platform – corporate process automation – in Hanoi, earlier this week. A press release said that the technology, akaBot is the FPT's third make-in-Vietnam platform. It is among 33 other platforms selected by MIC to introduce and sponsor media in the "Friday of Technology" programme, to serve the National Digital Transformation Programme to 2025, with the vision to 2030, which was approved by the Prime Minister in June. According to the development team, akaBot is a robotic process automation (RPA) solution for businesses with "virtual assistants" capable of simulating human manipulation, helping perform repetitive tasks in large numbers. With the core technology of RPA, akaBot is capable of integrating artificial intelligence (AI) and optical character recognition (OCR) technology to build comprehensive, non-invasive intelligent automation solutions, which can interact with all business software such as Word and Excel.