Microsoft Releases Speech Services

Useful capability for speech production from Azure/AI

Microsoft opens limited access to its neural text-to-speech AI in Engadget

AT&T and Warner Bros. used the tech to create an interactive version of Bugs Bunny.

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. ... 

Convert audio to text, perform speech translation and text-to-speech with the unified Speech services

Kris Holt, K. Holt|02.03.21  @krisholt in Engadget

Speech Services. https://azure.microsoft.com/en-us/services/cognitive-services/speech-services/

Google Makes more Speech Services Available

Impressive array of cognitive speech services, in 120 languages!   Now broadly available with demonstrations at the link.

Cloud Speech-to-Text

Speech-to-text conversion powered by machine learning and available for short-form or long-form audio.

Powerful speech recognition

Google Cloud Speech-to-Text enables developers to convert audio to text by applying powerful neural network models in an easy-to-use API. The API recognizes 120 languages and variants to support your global user base. You can enable voice command-and-control, transcribe audio from call centers, and more. It can process real-time streaming or prerecorded audio, using Google’s machine learning technology.

Some of the Betas in particular are indicative of future direction of capabilities:

Cloud Speech-to-Text features
Speech-to-text conversion powered by machine learning.
Automatic Speech Recognition
Automatic Speech Recognition (ASR) powered by deep learning neural networking to power your applications like voice search or speech transcription.
Global Vocabulary
Recognizes 120 languages and variants with an extensive vocabulary.
Phrase Hints
Speech recognition can be customized to a specific context by providing a set of words and phrases that are likely to be spoken. This is especially useful for adding custom words and names to the vocabulary and in voice-control use cases.
Real-time Streaming or Prerecorded Audio Support
Audio input can be streamed from an application’s microphone or sent from a prerecorded audio file (inline or through Google Cloud Storage). Multiple audio encodings are supported, including FLAC, AMR, PCMU, and Linear-16.
Auto-Detect Language BETA
When you need to support multilingual scenarios, you can now specify two to four language codes and Cloud Speech-to-Text will identify the correct language spoken and provide the transcript.
Noise Robustness
Handles noisy audio from many environments without requiring additional noise cancellation.
Inappropriate Content Filtering
Filter inappropriate content in text results for some languages.
Automatic Punctuation BETA
Accurately punctuates transcriptions (e.g., commas, question marks, and periods) with machine learning.
Model Selection BETA
Choose from a selection of four pre-built models: default, voice commands and search, phone calls, and video transcription.
Speaker Diarization BETA
Know who said what - you can now get automatic predictions about which of the speakers in a conversation spoke each utterance.
Multichannel Recognition BETA
In multiparticipant recordings where each participant is recorded in a separate channel (e.g., phone call with two channels or video conference with four channels), Cloud Speech-to-Text will recognize each channel separately and then annotate the transcripts so that they follow the same order as in real life.  .... " 

Baidu Mimicing Speech

Quite a need for Chinese regional accents.

Baidu’s text-to-speech system mimics a variety of accents ‘perfectly'
What used to take hours of neural net training now takes under 30 minutes. Timothy J. Seppala, @timseppala

Chinese tech giant Baidu's text-to-speech system, Deep Voice, is making a lot of progress toward sounding more human. The latest news about the tech are audio samples showcasing its ability to accurately portray differences in regional accents. ... " 

Alexa's New Speech Synthesis Language

I was just testing an App today, and was thinking about how the Alexa voice sounded natural and also robotic ... Now expect to see this to change as developers create applications that use it.   No public demonstrations exist yet of it.  I like to see how these continual developments will drive competition. Note this is NLP in the opposite direction we usually think of it, how speech is generated to create engagement or effect.

In the Verge:   by Ashley Carman 
  "  .....Amazon’s Alexa can now whisper, bleep out swear words, and change its pitch ... 
 Alexa can whisper, vary its speaking speed, and bleep out words. Developers can also add pauses, change the pronunciation of a word, spell a word out, add audio snippets, and insert special words and phrases into their skill. .... Amazon says this will allow Alexa to provide a “more natural voice experience ... "

Bots in the UnCanny Valley

The 'Uncanny Valley'  is a term used in the physical representation of people in android form,  as they get closer to reality we get uneasy. Now it has been suggested that the way bots chat is also reaching that.  I have conversed with bots where they seemed to have a personality.  Even the tone of Alexa when she says something like 'Good Night', can invoke an oddly social but uncanny feeling.  But a longer conversation,  which expects basic human performance, soon breaks that spell. So are we soon likely to combine image features and chat speech?  Will that be more or less disconcerting?

Chatbots Have Entered the Uncanny Valley 
The Atlantic,  Kaveh Waddell

The tendency for people to be repelled by increasingly humanoid and human-like robots may extend to chatbots and digital assistants as well. "The more human-like a system acts, the broader the expectations that people may have for it," says Carnegie Mellon University professor Justine Cassell. Modern chatbots use banter and humor, conversational speech, and parsing free-form questions and answers to coax users into engaging with them in more a human-like manner. "This creates a perception that if you say anything to this bot, it should respond to it," says Autodesk engineer Nikhil Mane. This makes for situations in which user requests exceed the bot's limitations, and the subsequent errors serve to remind users of the assistant's artificiality. Mane says a better approach for bots is to make users aware of their constraints, such as prompting  ... " 

Amazon Rolls Out Lex Chatbot Tools

Amazon rolls out conversational Chatbot interface 'Lex'.  Am looking for better yet conversational and memory capabilities for Lex.   Good technical overview at the link.

Amazon Lex – Build Conversational Voice & Text Interfaces  by Jeff Barr  
While computers that talk are great, computers that listen and respond are even better! If you have used an Amazon Echo, you know how simple, useful, and powerful the Alexa-powered interaction model can be.

Today we are making the same deep learning technologies (ASR – Automatic Speech Recognition NLU – Natural Language Understanding) that power Amazon Alexa available to you for use in your own conversational applications. You can use Amazon Lex to build chatbots and other types of web & mobile applications that support engaging, lifelike interactions. Your bots can provide information, power your application, streamline work activities, or provide a control mechanism for robots, drones, and toys.

Amazon Lex is designed to let you get going quickly. You start out by designing your conversation in the Lex Console, providing Lex with some sample phrases that are used to build a natural language model. Then you publish your Amazon Lex bot and let it process text or voice conversations with your users. Amazon Lex is a fully-managed service so you don’t need to spend time setting up, managing, or scaling any infrastructure.  ... " 

HPC and Deep Learning by Baidu

That Baidu opening up here is interesting, read the full article at the link.  Openness of Baidu is intriguing.  See an intro to OpenMPI.

HPC Technique Propels Deep Learning at Scale   By HPC Wire 

Baidu's Silicon Valley Artificial Intelligence Lab (SVAIL) has released a modified implementation of the ring all-reduce OpenMPI algorithm for the deep-learning community, which will enable faster training of neural networks across graphical-processing unit (GPU) nodes.

Unlike the OpenMPI version, the SVAIL modification avoids making extraneous copies between the central processing unit (CPU) and the GPU.

Although commonplace in high-performance computing, the technique has been underused within AI and deep learning, according to Baidu. Compared with using a single GPU, the ring all-reduce algorithm is about 31 times faster at 40 GPUs.

The algorithm has enabled the SVAIL team to get linear GPU scaling up to 128 GPUs and to parallelize the training of Deep Speech 2, its speech-recognition mode.

Two years after the approach was initially developed, the researchers have issued two non-proprietary implementations, one for TensorFlow and one for more general applications. ... " 

Voice Authorization

I have not had any issues like this, but it makes sense.  Even better would be the ability to differentiate between household members.   No current assistant does this. Must be hard?

Voice-checking device stops hackers hijacking your Siri or Alexa  By Matt Reynolds

Does your digital sistant know who it’s talking to? A wearable device prototype could let voice-controlled assistants like Apple’s Siri or Amazon’s Alexa recognise their owner so they don’t take orders from anyone else.

The VAuth device, developed at the University of Michigan in Ann Arbor, uses an accelerometer hidden in a pair of glasses or earphones or worn around the neck. The accelerometer measures vibrations created as the wearer speaks. An algorithm then compares those vibrations with the audio signal received by the digital assistant. If the vibrations and the audio match, then the voice command is received as normal. If not, the assistant is blocked from responding.

Computer scientists Huan Feng and Kassem Fawaz at the University of Michigan  tested VAuth with 18 people saying 30 different commands. It was able to match speech vibrations with audio signals 97 per cent of the time, and did not act on commands issued by someone other than the device owner. Although they tested VAuth only with Android assistant Google Now, Feng and Fawaz are confident it could work with other voice assistants if their manufacturers allowed it.  ....  " 

Enabling Voice Control in Simple IoT

From MIT News  in CACM

" ... In anticipation of the age of voice-controlled electronics, MIT researchers have built a low-power chip specialized for automatic speech recognition. Whereas a cellphone running speech-recognition software might require about 1 watt of power, the new chip requires between 0.2 and 10 milliwatts, depending on the number of words it has to recognize. ... 

In a real-world application, that probably translates to a power savings of 90 to 99 percent, which could make voice control practical for relatively simple electronic devices. That includes power-constrained devices that have to harvest energy from their environments or go months between battery charges. Such devices form the technological backbone of what's called the "Internet of Things,"or IoT, which refers to the idea that vehicles, appliances, civil-engineering structures, manufacturing equipment, and even livestock will soon have sensors that report information directly to networked servers, aiding with maintenance and the coordination of tasks. ... "

SoundHound

Had heard of these folks, in particular of their work with Samsung.   Potentially drives voice to assistants and chatbot capabilities.    Competitive to Alexa and Google Home directions.

SoundHound raises $75M to create a voice-controlled future   by Maria Deutscher
For SoundHound Inc., developing a wildly popular app with hundreds of millions of downloads was just the first step towards success.

The Santa Clara-based startup, which has carved out a niche by helping mobile users identify unfamiliar songs, raised $75 million this week to launch the next stage of its plan. SoundHound is looking to establish the sophisticated artificial intelligence at the core of its app as the industry’s leading speech recognition platform. The firm is targeting not only music and other forms of entertainment but also the growing number of voice-controlled technologies that are entering the market.

This trend is driven in large part by the rapid adoption of virtual assistants that are automating menial administrative tasks previously carried out by humans. Some banks, for instance, rely on the technology to handle customer support, while other companies employ chatbots to streamline their cybersecurity operations. More and more of these bots are being equipped with speech recognition algorithms that enable them to take voice commands. ... " 

Q&A on PowerAI

For research I took a look at the PowerAI offering.  See the complete article:

Putting the “AI” in PowerAI
A Q&A with IBM Research’s Hillery Hunter

IBM’s latest Power servers come with an AI twist. Optimized for deep learning, a new so-called PowerAI toolkit will “help train the systems to think and learn in a more human-like way, at a faster pace,” as announced at SC16, the International Conference for High Performance Computing, Networking, Storage and Analysis.

I spoke with Hillery Hunter, IBM Research’s director of Systems Acceleration and Memory and Memory Strategist, about her team’s contribution to the software behind the world’s fastest deep learning servers.

How was the AI added, and what does that mean for how a Power system functions?
 Today’s launch represents our first public offering in hardware-software co-optimization for deep learning. Researchers have worked closely with IBM’s systems engineers to create code that is optimized to the Power S822LC (IBM’s highest-performing OpenPOWER server).

We used the S822LC’s unique design – including high-bandwidth NVLink interconnect, not just between pairs of GPUs, but also between the GPUs and the CPUs – to deliver higher deep learning training performance.

This means that the co-designed PowerAI hardware and software can build learned models from images, speech, or other media in less time than prior generations of hardware and software. Deep learning training time is a key metric for developer productivity in this domain. It enables innovation at a faster pace, as developers can invent and try out many new models, parameter settings, and data sets.    .... " 

Tensorflow and PowerAI

Great to see integration of key AI software, lets also start to see excellent examples of their use with supporting data architecture.

IBM brings Google's AI tools to its Power hardware

IBM is integrating the open-souce TensorFlow into its PowerAI toolkit
By Agam Shah  

Google has cool technology to recognize images and speech, and IBM's hardware can diagnose diseases. ... Combine the two, and you get a powerful computer with serious brains.

IBM is merging Google's artificial intelligence tools with its own cognitive computing technologies, allowing deep-learning systems to more accurately find answers to complex questions or recognize images or voices.

Google's open-source TensorFlow machine-learning tools are being packed into IBM's PowerAI, which is a toolkit for computer learning. The two can be combined to improve machine learning on IBM's Power servers.

A computer learns as more data fed into its system, much like how a human learns. PowerAI and TensorFlow tools can help track patterns and classify data, and spit out approximate answers to queries. The answers will be more accurate as a computer learns more.  .... " 

Fielding a Hundred Thousand Questions

Via O'Reilly.   We did something similar when creating our Extempo bot system, but only included a few hundred questions.  Nice to see Microsoft pushing research with this.    How many parameterized questions does a human ask in normal conversation?   How does this relate to context? To get to useful answers?  Good, largly non-technical article.  If you are at all interested in bot like interaction, worth a look.  Thinking of feeding selected elements of this through some current assistants.

100,000 questions (and answers)
Microsoft has released a set of 100,000 questions and answers that you can use to create AI systems that can read and answer questions as well as a human. The team behind the dataset hopes that releasing it will spur the kind of breakthroughs in machine reading that are already happening in image and speech recognition.  ... " 

Shouting at Cortana and the Value of Voice

Though a keen observer and frequent user of voice activation in the smart home, I rarely talk to my otherwise much in use laptop.  I was at first impressed with aspects of MS Cortana, but soon found it boring.  It did not support what I wanted to do, when I wanted to do it.   So I stopped talking and forgot that Cortana was there.   Now word that you can soon talk to it from further away, and activate it more readily.  Is that enough?  Am using Windows 10, so I am all set up to test the idea.

Intel, Microsoft going long distance with Cortana on Windows 10 PCs
You'll also be able to shout 'Hey Cortana' to wake up PCs from standby
By Agam Shah  

Intel and Microsoft are working on technology allowing you to shout out commands to Cortana or activate a Windows 10 PC from sleep mode without being all that close to it.

The chipmaker is working with Microsoft to add "far-field speech recognition" technology, where the user can shout out Cortana commands to a Windows PC from longer distances.

"Soon, you’ll be able to speak to your PC from a distance and access all of your information on the device and in the cloud," said Navin Shenoy, senior vice president and general manager for the Client Computing Group for Intel, in a blog entry this week.    ... " 

Amazon Delivers Serious AI Services

   More tools being added to the box.  Previously some from IBM and Google.  Can remarkable AI be far behind?  In Computerworld.

AWS comes out swinging with A.I. services
Trying to entice enterprises and catch up with cloud rivals, AWS releases three artificial intelligence services   By Sharon Gaudin  

 .... At its re:Invent conference today, Amazon Web Services CEO Andy Jassy announced three artificial intelligence (A.I.) services that will be available to enterprise users this year, with more expected in 2017.

The three services being rolled out are Amazon Rekognition for image recognition; Amazon Polly for text-to-speech services; and Amazon Lex, the technology inside its smart device Alexa, offering speech recognition services.

Now enterprise AWS customers can use A.I. to, for example, search for images that show a mountain next to a lake, or a city with specific architecture. ... " 

via Walter Riker. 

Using IBM Watson in Business

This oversimplifies the task of using a capability like IBM/Watson, but is still a good introduction. Non technical, suggestive as to the possibilities.   Also an excellent list of companies using Watson.  Pointers to more detail.  By William Vorhies in DSC:

" ... In our recent reviews of historical Watson and the modern Watson of today we concluded that IBM’s Watson Group may have the first or at least the current strongest comprehensive AI platform. 

 This is the first time that we know of that all three elements of AI have been brought together in a single user friendly platform: image processing, text and speech processing, and knowledge retrieval. 

This is not so much a platform for data scientist to use to expand the capabilities of AI as it is a platform for business users (with the aid of data scientists) to exploit the capabilities of modern AI by building new products and services.

To wrap up this review of Watson, we wanted to provide some thought-starters on what new services or even new businesses you might build on Watson.

Oh, and regarding new businesses, did we mention that developers who join the Watson Ecosystem are eligible to become a Watson “partner” with a shot at the $100 Million funding IBM is making available to startups plus support and access from IBM business and technology advisors. ... " 

New Chip Usage to Power AI

Re-using chip designs to perform the methods used for AI applications.  Now new chip designs are in play.

Intel Looks to a New Chip to Power the Coming Age of AI
Microsoft Researchers recently built an artificially intelligent system that seems to recognize conversational speech as effectively as a human. Yes, this research comes with caveats, but it’s part of a very real and very rapid leap in artificial intelligence over the past several years, a leap driven by deep neural networks.

These sweepingly complex algorithms can teach themselves very particular tasks by analyzing vast amounts of data. Microsoft’s system learned to recognize words by looking for patterns in old tech support calls. But it’s not just the algorithms that are driving the recent revolution in AI. It’s also the hardware behind these algorithms. Microsoft’s speech rec system relies on large farms of GPU processors, chips that were originally designed for rendering graphics but have proven remarkably adept at running artificial intelligence models. .... 

Internet giants like Microsoft, Google, Facebook, and Baidu typically train their deep neural nets using GPUs. But they’re moving towards other, more specialized chips that can help accelerate not just the training but the execution of these systems. Google recently built its own AI processor. IBM is building another.

So Intel, the world’s largest chip maker, is doing the same. Yesterday, the company unveiled a new AI processor called Nervana, saying it plans on testing prototypes by the middle of next year, and if all goes well, a finished chip will come to market by the end of 2017. At the moment, the market for AI chips is dominated by nVidia, the primary supplier of GPUs.  ... " 

Surface Dial Examined

I have been a long time experimenter with alternative interfaces.  From mice, to feet driven mice, to touch pens, speech and neural interaction and more.  So I was interested when I saw the 'Surface Dial', being used for another channel of interaction on the Microsoft Surface.  There are other similar devices out there as well.  Most important, what software and other hardware works with the Dial? How can this be used efficiently with other interactions, like voice and mice and touch?  What does it mean for an integrated assistant like Cortana?  It could lead to the ultimate need for 'hands free' interaction, since it engages another hand.   Verge talks software and the Dial system.

Neural Brain to Text Progress

Remember this topic being brought up for a long time, and often claims that is was close to happening.  Have now been involved in the spoken interface for some time, a reliable thought interface sounds very appealing.   Considerable detail in the full article about the challenge.

Front. Neurosci., 27 September 2016  
Automatic Speech Recognition from Neural Signals: A Focused Review

Christian Herff* and Tanja Schultz
Cognitive Systems Lab, Department for Mathematics and Computer Science, University of Bremen, Bremen, Germany

Speech interfaces have become widely accepted and are nowadays integrated in various real-life applications and devices. They have become a part of our daily life. However, speech interfaces presume the ability to produce intelligible speech, which might be impossible due to either loud environments, bothering bystanders or incapabilities to produce speech (i.e., patients suffering from locked-in syndrome). For these reasons it would be highly desirable to not speak but to simply envision oneself to say words or sentences. Interfaces based on imagined speech would enable fast and natural communication without the need for audible speech and would give a voice to otherwise mute people.

 This focused review analyzes the potential of different brain imaging techniques to recognize speech from neural signals by applying Automatic Speech Recognition technology. We argue that modalities based on metabolic processes, such as functional Near Infrared Spectroscopy and functional Magnetic Resonance Imaging, are less suited for Automatic Speech Recognition from neural signals due to low temporal resolution but are very useful for the investigation of the underlying neural mechanisms involved in speech processes. In contrast, electrophysiologic activity is fast enough to capture speech processes and is therefor better suited for ASR. Our experimental results indicate the potential of these signals for speech recognition from neural data with a focus on invasively measured brain activity (electrocorticography). As a first example of Automatic Speech Recognition techniques used from neural signals, we discuss the Brain-to-text system. ... " 

Microsoft Announces Speech Recognition Breakthrough

Continued advances in Cognitive interaction.  While speech recognition is good today,  making it better will remove one barrier to interacting nimbly with spoken language.  'Understanding' semantically, with common sense implied,  is still not easy.

In SiliconAngle: 
In an announcement that could be the death knell for stenographers everywhere, Microsoft Corp says that it has made a major breakthrough in the field of speech recognition. According to the company, it has developed an artificial intelligence that is capable of understanding conversational speech at a rate comparable to real humans.  

 In a research paper published through Cornell University on Monday, Microsoft’s Artificial Intelligence and Research team outlined the test they used to determine the effectiveness of their new system. According to the researchers, they tested their AI with the NIST 2000, an evaluation created by the National Institute of Standards and Technology that is specifically designed to determine the accuracy of speech recognition software. ... "