Quality Speech Dataset Recognition In Order To Ignore Errors
\
With new devices that can be activated by voice each week, one could think that we're at an end of the road for technology for speech recognition. Yet, a recent Bloomberg article asserts that, while the technology of speech recognition has seen huge advances in recent times but the way it is implemented to the process of Speech Data Collection has hindered it from reaching a level where it can replace how people communicate with devices. The public has embraced the idea of devices that can be activated by voice with enthusiasm, however the actual experience still has potential for improvement. What's holding this technology behind?
More data = better performance
As per the author's report, what's required to enhance devices' ability to comprehend and interact to users are terabytes spoken human voice data comprising various accents, languages, and dialects that can enhance the ability to understand conversations that the gadgets have.
Recent advancements in speech engines are due to a form of artificial intelligence known as neural networks which are able to learn and adapt with time without precisely programmed. In a loose way, they are modeled after our brains and brains computer systems can be trained to understand the world around us, and perform better when they have more AI Training Data. Andrew Ng, Baidu's chief scientist, states "The more data we store into our systems, the more efficient it is. This is the reason why speech is a capital-intensive procedure; not a lot of businesses have this kind of information ."
It's all about quality and quantity.
While the amount of data is essential but the quality of data is crucial to optimize machines learning techniques. "Quality" in this case refers to how well the data is suited to the purpose. For instance, if a voice recognition system is developed for use in cars it is necessary for the data to be collected inside a car to get the best results, taking into consideration all the usual background noises that the engine will 'hear'.
While it's tempting using "off-the-shelf" data, or to gather the data with random methods, it's more effective long-term to gather data specifically for its intended use.
This is also true when creating global speech recognition software. Human speech data is nuanced, inflected, and full of biases based on culture. Data collection needs to be done across a wide range of languages, geographic accents and locations to decrease errors and boost performance.
What happens when Speech Recognition Goes Wrong
Speech recognition that is automatic (ASR) is one of the things we use each day at GTS. Accuracy in speech recognition is one of the things we are proud of by helping our customers achieve and we're confident that those efforts are appreciated all over the world as more and people are using speech recognition on their smartphones as well as on their laptops, or even in the home. Personal assistants from digital technology are available at our fingertips and are asked to schedule reminders, answer messages or emails, or to look up us and recommend a good place to take us for a meal.
All well and good however, even the most advanced voice recognition technology has trouble in achieving 100 percent accuracy. When errors occur it are often glaring or even amusing.
1.What kind of errors occur?
A device that recognizes speech will usually produce the words it hears depending on the information it has received as that's what they're made to accomplish. However, choosing which word that it's heard can be a difficult task as there are handful of factors that can cause users to be confused.
2.Making the wrong assumption about the word
This is, naturally, the main issue. Natural language software does not have the ability to form complete plausible sentences. There are a myriad of possible misinterpretations that could be similar, but they don't make much sense as a whole sentence:
3.Listening to things that aren't the words you were using
If you pass someone and you're speaking in a loud voice or you cough during a phrase the computer isn't capable of determine which part of the audio you were speaking and which were a different part that is playing. It could result in things such as a phone taking a transcription while they were practicing using the tuba.
4.What's the deal there?
What is the reason these well-trained algorithms making errors that any human person would find hilarious?
5.What can people do when their devices fail?
If things go wrong in the accuracy of speech recognition the chances are that they will remain in the wrong direction. The general public is cautious when speaking to a virtual friend at even the most tense of times. it's not difficult to undermine that trust! Once an error is made there are all kinds of bizarre things to clarify themselves.
Certain people slow down. Some people might over-pronounce their words and ensure the Ts and Ks are as clear as they possible. Others will attempt to mimic the accent they believe the computer can most easily understand, and do their best impression of queen Elizabeth II or of Ira Glass.
But here's the problem While these methods may assist if you're talking to an inexperienced tourist or people on a poor call, these techniques do not assist computers at all! In reality, the more we go from natural, connected speech (the kind found in the recordings that were used to train the recognizer) the more difficult the situation will become, and the downward spiral will continue.
Comments
Post a Comment