Showing posts with label speech recognition technology. Show all posts
Showing posts with label speech recognition technology. Show all posts

Tuesday, March 30, 2010

Nuance Shutters SpinVox Consumer Service

It's been just over 3 months since Nuance acquired SpinVox, saving the company from a death spiral. Since the acquisition a string of news stories have surfaced about the financial shenanigans that went on at SpinVox prior to Nuance's purchase. It's a shame that Spinvox fell to such lows. They had a great idea and were developing credible technological solutions.

A recent and not too surprising post on SpinVox's website indicates that they will discontinue their consumer offerings allowing them to focus on their carrier and network operator business.

A Twitter post from SpinVox stated: "We regret to inform you that SpinVox is no longer supporting individual user accounts. Your account will expire in 7 days. "

No word yet on what they intend to do with consumer customers of Jott, the Seattle based competitor of SpinVox that Nuance also purchased.

Friday, July 24, 2009

On the BBC news story bashing SpinVox

I read with interest this morning a BBC story about SpinVox which suggests that the majority of messages on it's platform have been heard and transcribed by call centre staff in South Africa and the Philippines rather than being transcribed into text using speech recognition technology.

The article goes on to say that messages appear to have been read by workers outside of the European Union raises questions about the firm's data protection policy. SpinVox's entry on the UK Data Protection Register says it does not transfer anything outside the European Economic Area.

Anyone with a working knowledge of the voice mail to text transcription industry (which includes other vendors like Jott Networks and Google Voice) understands that no speech recognition process available today can achieve perfectly accurate automated transcriptions for large numbers of voice mail messages from thousands of different callers and the wide variety of audio quality typical of phone calls, especially those like poor cellular connections.

Today, almost everyone working in this space uses a combination of speech recognition technologies and human (read: caller center agent) based quality assurance (q/a) to obtain transcriptions of a usable quality. The human touch adds two elements: first, it can edit out errors from the automated transcription process and secondly, the markup data from the human q/a agents can be used to further refine the recognition process.

In the rare cases where no human q/a is used before delivering the transcription to the end users, the quality of the transcription almost always suffers. By example, has anyone seen "Great" transcription yet from Google Voice?

Unfortunetly, the economic model playing out in this industry forces this q/a work to off-shore or third world call centers.

The BBC story is important in it's discussion of the data security issues. So far, none of these services has provided sufficent details about the processes they use to assure data security and it does appear on the surface that SpinVox may be violating the EU Data Protection Policy that it's committed to. To quote Ross Perot: "The nut's in the detail". We've not yet seen enough detail to know much about the "nut".

I've used Jott, Google Voice and SpinVox myself (in fact I currently use SpinVox on my cellular voice mail) and I've found all to be useful but none to be superbly accurate in their transcriptions. However the services with human based q/a have faired much better.

What's your experience been? Are you concerned about the security of your message content when using these services for voicemail transcription? I'd be interested in hearing your comments!

Wednesday, April 29, 2009

TellMe Rolls Out New Features

Tellme®, a subsidiary that Microsoft® purchased two years ago has announced a Spring Release of new features and product improvements that tighten the integration between TellMe & it's parent and take advantage of the deep technical skills from both TellMe and Microsoft's speech research group.

The new release offers improvements focused on:
  • Positioning TellMe as a service offering in the emerging Cloud Computing arena
  • Improving core speech recognition quality
  • Adding new multi-slot recognition capability
  • Reducing telecom costs

The enhancements include a new VoIP offering through a partnership with Global Crossing, a new TTS voice which leverages the Microsoft Text-to-Speech (TTS) engine, new acoustic models, phonetic dictionaries and grammar products that increase the accuracy, new multi-slot dialog capability which will allow callers a more natural conversational experience while keeping recognition rates high and new mobile services, including the Windows Mobile 6.5 application. They will now be able to offer not only Nuance and IBM recognition engines but also add Microsoft's engine.

I've hyperlinked to the announcement press release in the opening paragraph so you can take a look at the changes in more detail.

TellMe has always been a strong hosting company and these additions will only further strengthen their position in the market.

Monday, April 27, 2009

The Clouds are Gathering - Speech 2.0

When I lived in north Texas, it was easy to see a storm coming – the thunderheads gathered on the horizon to the west and you could see them building for hours before they arrived.

After eComm 2009: Emerging Communications Conference I saw the same kind of storm clouds gathering in the west. This time rather than bringing wind, rain and lightening I believe they mark a significant shift in the evolution of speech recognition – from hosted VXML based platforms to Telephony “in the cloud” or what’s being called Speech 2.0.

This shift builds on the trend towards cloud computing; the hosting of applications and more importantly services in the cloud and provides a number of potential benefits including:


  • Reduced or no front end capital investment
  • Reduced operating expenses and easier upgrade/scalability
  • Rapid speed to market deployment for new application
  • Easier integration with and synchronization with existing web self-service
  • Access to best-of-breed technologies
  • Superior network reliability and redundancy
  • Gigabyte interfaces for fast reliable operations
  • Scalability to meet high or rapidly changing call volumes
  • Compatibility with existing telephone and Web infrastructure
Several specific Speech 2.0 platforms were presented at eComm, each with its own twist on features, supported interfaces and hosting options. All of them offer some form of free developer account so that you can sign up and try them yourself.

The platforms that I took note of were:

Tropo – Voxeo introduced Tropo, an in-the-cloud development platform that lets users create and deploy speech and telephony applications using a simple API (application programming interface). The API support application development in Groovy, JavaScript, PHP, Python, and Ruby and is designed as an alternative to the standard XML- and VoiceXML-based platforms that have become so common in the last few years. Applications can incorporate inbound calling via the public switched telephone network, Session Initiation Protocol (SIP), Skype, and iNum, while also providing appropriate connections for outbound calling. Capabilities include robust call control, playing and recording audio, touchtone entry, speech recognition, text-to-speech, and mashups with Web services. Planned application capabilities include call recording, conferencing, and Web services.

Twilio - Twilio provides an in-cloud API for voice communications that leverages existing web development skills, resources and infrastructure. Designed to enable web applications to be able to interact with phone callers, Twilio allows you use your existing web development skills, existing code, existing servers, existing databases and existing karma to solve these problems quickly without the need to learn some foreign telecom programming languages, or set up an entire stack of PBX software. Twilio provides the infrastructure; you provide the business logic via HTTP. Currently PHP, REST.

IfByPhone – IfByPhone is a hosted voice application and platform company with a simplified approach to the deployment of stand-alone and web-integrated voice services for small and medium sized businesses (SMB). Through a combination of telephony and web services IfByPhone offers you prebuilt applications or a programmable API which enables you to create inbound or outbound calls or other IVR functionality. The configuration and deployment tools look and feel just like Web applications, and require no previous knowledge of telephony programming or terminology.

In the spirit of full disclosure, I use a mash-up of IfByPhone’s applications, a “Click to Call” button on my own website which allows callers to connect to me directly from my website. It prompts the caller for their phone number, places a call to them, and then invokes a “Find Me” application that places simultaneous calls to me at several possible numbers and bridges which ever number I answer on to the caller.

Jaduka – Jaduka provides a SOAP-based Web Services interface which enables companies to easily blend voice into their workflow activities Jaduka's Web services API makes adding the benefits of voice communication to enterprise applications as easy as constructing a mash-up.

I’m sure there are other platforms that I’ve missed. If I’ve overlooked you, drop me a note.

These cloud-based telephony providers start with a hosted platform, accessible via the open Internet and provide a an API whose premise is to enable developers not familiar with speech recognition or telephony to incorporate a speech based channel into their existing application infrastructure without having to learn a great deal of domain specific skills or development languages, like VXML or CCXML that are typically used today to build voice enabled applications. Each varies slightly but in general the primary interfaces are built on common web based development languages and protocols such as REST, SOAP, PHP, Groovy, JavaScript, Python, and Ruby. In addition, some of these platforms have also developed more complete applets or small applications which can be used right of the box or with a nominal amount of configuration.

In theory, this approach makes it easy for developers to add voice interfaces to existing applications. But as your mother reminded you when you were are child: “just because you can does not mean you should”. Speech enabled applications have taken a long time to reach the main stream success. There are many reasons for this, one of which was the need to develop experience in what’s required to provide users with an interface that works well for the caller. There is as much art to this as there is science and technology. As with many endeavors, our first attempts often fail to meet expectations. Developers using these cloud based platforms will still need to design and implement good users interface practices for voice which are much different than those applied in visual application interfaces. Much has been written about this, so I’ll leave that topic to another conversation.

This evolution in platforms when added to the strong uptake in hosted platforms is sure to have significant impact on the business models in the speech recognition and telephony industries. With the concentration of fewer and larger consumers of the core technology, pricing and volume of sales with no doubt change for firms like Nuance and the various IVR hardware vendors.

Over the next month, I'll explore these new cloud based platform individually by building a application on each and sharing my experience and results with you. I'll wrap up this series of posts with some kind of comparison post giving you my perspective on the pros and cons of each. Stay tuned!

If you're already using or experimenting with one of these cloud based telephony platforms I'd love to hear from you about your experience as well.

Thursday, April 09, 2009

Collecting SLM Data in an existing self service application

I had the opportunity to call Comcast (my internet service provider) this week about a problem with my cable modem.

The call was answered by the usual self-service application that asked me a series of questions (a directed dialog style) including my home phone number and then if I was calling about my cable TV service, Internet service or phone service. At that point the application took an unexpected change of direction by asking me to state why I was calling. After I said "Internet not working" the application informed me that it collected that information for a future application enhancement then went back to the normal directed dialog style with 5 or 6 more questions. No doubt, the data collected will be pared with my ultimate destination and purpose in the call to help develop a natural language component.

I immediately thought of a post by Phillip Hunter this week on his Design-Outloud blog and the related article in SpeechTech Magazine: Is it Natural? I'd encourage you to read both, especially Phillip's blog post. These will give you a good overview of the challenges to this approach.

It's obvious that Comcast is gathering data to add some kind of "Natural Language" capability to their customer self service application. This would imply that Comcast has a very high call volume for this application, as it typically costs a large 6 figure number to build a Natural Language call steering application based on SLM (statistical language modeling). It's often difficult to justify the expense of building and maintaining this kind of approach.

While this approach is often very productive and cost justified for high volume callers, an equally workable approach with a slightly narrower focus can be developed using a typical SRGS grammar for a much lower cost, using an technique known as grammatical inference.

Essentially grammatical inference tools rely on artificial intelligence to build your SRGS and GSL grammars automatically based on example utterances. Just as you build an SLM grammar using caller utterances, with grammatical inference, you feed the utterances into a ‘grammar learning tool’ which outputs a set of grammar rules in whatever format you require (e.g. GSL or SRGS). The grammar learner has a fundamental knowledge of the language that you are building the grammar for (e.g. English) and combines this with your utterances to produce a set of grammar rules. Unlike an SLM approach, grammatical inference allows you to build a usable grammar with only a very small number of sample utterances (‘tens of utterances’ rather than ‘tens of thousands’). Of course, if more training data is available, you can feed the grammar learner as much as you like. One such tool that I've used is offered by Inference Communications Pty. Ltd. I'm sure others are out there waiting to be found.

When working with my clients, I often find that major speech vendors jump straight to a recommendation of Natural Language (with glimmers of large licensing and professional services fees in their eyes) when a more conservative and less expensive approach "may" work just as well.

Regardless of the direction you choice, consult with a skilled voice interaction designer who can help you parse through the pros & cons for each approach and choose the right one for you and your callers.

P.S. Also worthy of mentioning this week is a new web site from The Association for Voice Interaction Design. If you have anything more than a passing interested in voice interaction design, they have some great references to other blogs, websites and publications.

Thursday, April 02, 2009

Despite being a skeptic . . I'm impressed

You would think that since I've made a career of working with speech recognition technology that I would be wildly enthusiastic about it and just like a carpenter with a hammer would see every problem as a nail. Not so with me!

I got into this technology reluctantly and have remained largely skeptical ever since. It's just in the past few years I have become truly comfortable in recommending voice recognition technology as a viable and solid technology to be used in business solutions. Even then, I'm careful and conservative in when and how I make that recommendation. I have to be 100 percent confident that the proposed solution can succeed and will prove successful both in the mind of the user and the business.

Along the way I've tried, sampled and tested almost everything offering including the desktop recognition packages from IBM's ViaVoice, Nuance's Dragon Naturally Speaking and most recently the speech recognition included in Microsoft's Windows Vista. After every one of these trials, I've uninstalled or disabled the recognition package and gone back to my keyboard. While they worked to a greater or lesser degree they never "excited me" and often were more trouble than they were worth.

Despite previous mediocrity, I keep trying new versions or products. This week I finally got around to installing Nuance's Dragon Naturally Speaking 10, after it lay on my desk for month's.

The installation process was quick and easy with no complications and once installed the product walked me through a basic set process to set and calibrate the audio input source (I have multiple devices with microphones attached to my Dell Vostro with Windows Vista) and complete a short training exercise by reading a Dave Barry column so that Naturally Speaking could create a user profile for me. All done in less than one half hour from start to finish, including time to download and install the Microsoft Visual C++ Redistributable.

For those of you who know me in person, you'll no doubt remember that I have a distinct southern accent, having grown up near Memphis, TN. One of the user configuration options was a choice of languages that included General US or Southern US. I resisted temptation and chose General just to test Naturally Speaking a bit more.

Now after a few days of use, I have to admit amazement! The recognition is incredibly accurate and quick. I've not typed an instant message or email so far this week. All have been dictated using Nuance's Dragon Naturally Speaking 10. In fact this entire blog entry was done using speech dictation.

I've only experienced two problems:

1) Some issues with product names or proper names like Vostro, which require correct.

2) I think I need a more highly directional microphone for my headset. I listen to my local public radio station in the background while at my desk and if I'm quite without telling the software to stop listening it will start transcribing the conversation on KUOW 94.9 that's playing in the background. This might make it difficult to use Dragon in a loud office or call enter environment.

That's my two cents worth . . . and if you're interested, I have a copy of Dragon Naturally Speaking 9.0 that I never got around to installing. It's for sale on eBay!

The Good, The Bad and The Ugly - Real World Examples of Speech Enabled Self-Service

When discussing issues related to a particular approach or dialog in a speech recognition based application with my clients, I find it's very helpful to have them call and interact with examples in a real application. Often interacting with a real world example, illuminates their thinking more clearly than any intellectual discussion. It makes the point I'm trying to illustrate very visceral for them, having experienced the pain or the pleasure.

I keep a document with several dozen of these real world examples which I give my clients as a reference. It's time for me to refresh the content for this document and I thought I'd ask for your input. I've put together a survey to collect this information. Click Here to take survey

Once I've gathered the results and updated my reference document I'll post a link to it here and discuss some of the most note worthy examples.

Friday, October 03, 2008

California Bans Texting and Driving

We all know in our "gut" that texting on our cell phones while driving or conducting any other task that requires high concentration and motor skills is seriously distracting. A good example of this is the recent fatal train accident in southern California. The NTSB (National Transportation Safety Board) has determined that the engineer of the train that caused the accident was sending text messages from his cell phone during the last few minutes before the accident and the belief is that this may have contributed to the accident.

Last fall at Nuance Communication's CONVERSATIONS Conference, one of the keynote sessions involved a demonstration of just how dangerous and distracting "texting" was. You can see a video of this demonstration on YouTube at Amazing Race: Distracted Driving.

This past week California's Governor, Arnold Schwarzenegger said "Hasta la vista" to texting while driving and terminated a loophole in California's vehicle code that banned drivers from talking on cell phones while driving without a hands-free device but let them communicate via text messages. The Associated Press reported that the governor signed the law which will take effect on January 1, 2009. My home state of Washington had already banned text messaging while driving.

Given the usefulness of these devices it's difficult to image that we'll break our addiction to the BlackBerry or iPhone -- they don't call it a "CrackBerry" without a reason. That said, speech recognition technology offers a straight forward way to improve safety in the way we use these devices without taking away their convenience. Several services and embedded applications have already reached the market which address this problem. One example that I've blogged about before is Jott Networks. Jott allows you to dial their service and then send notes to yourself and others entirely using voice recognition based navigation and dictation. Others companies with similar offerings include SpinVox and SimulSays. When combined with voice based dialing and hands free access, these services remove most of the physical contact required to interact with mobile devices, eliminating much of the distraction that occurs when you interact via the keyboard.

This lesson can be carried over to the customer service world. Many IVR based self service applications in use today require serious use of the DTMF keypad for entry of things like account numbers, choices from lists, ticker symbols, etc. and pose the same risk to drivers as other kinds of text messaging. Given the prevalence in the use of cell phones these days, I think a strong case can be made for speech enabling applications which have complex DTMF (touch tone) entry requirements simply as a safety step for callers and to avoid potential legal issues which may become a problem as more and more states impose similar bans on non-hands-free use of cell phones.

Monday, June 30, 2008

Hands Off - it's the law!

Beginning this week drivers in California and Washington join those in a list of other states who can't use their cell phones while driving to talk or send text messages unless they're using them in a hands free mode. In my home state of Washington, drivers who read and compose text messages or talk on a cell phone without a hands-free device could face a $101 ticket. The text-messaging ban took effect Jan. 1; the cell-phone law will be enforced starting in July 1st. Drivers are exempt in some situations, including emergencies, and neither offense will be enough to get a driver pulled over by the police.

Several of my clients who still have complex DTMF applications in service with high cell phone caller populatations which require lots of DTMF entry are moving rapidly to migrate those applications to speech recogntion based self-service to avoid any potential liability issues and to make sure that the applications remain available during the periods when callers actually want to use them, namely while driving or otherwise in transit.

A stunning 80 percent of mobile phone owners talk while driving, according to a recent survey by the Nationwide Mutual Insurance Company. It's a major distraction—some have even equated using a cell phone behind the wheel with driving under the influence, since reaction times can be slowed during a call.

According to the National Highway Transportation Safety Administration, there are 115 road fatalities each day in the United States and distracted driving causes 80 percent of road accidents.

I've been making the recommendation to my clients that these facts alone justify the move from DTMF to speech recognition based functionality for any self service application with modest to high cellular caller populations and more than the simplest of input requirements.

I'm curious if anyone else is seeing this move to hands free cellular use as much of a driver for moving DTMF applications to speech recognition based self service or if anyone else has had the issue of liability raised.