Ring-tailed lemur

 

Nice to meet you! I'm a ring-tailed lemur.

I joined a company called Advanced Media Co., Ltd. as a new graduate and worked on the voice recognition API "(AmiVoice Cloud Platform) " is involved in the product.

I would be happy if my message could be of help to as many people as possible.

 

Is voice recognition weak at "proper nouns"?

Now, I would like to talk about the main topic of this article, "How to recognize proper nouns using voice recognition." The reason I decided to write this article isVoice recognition has the characteristic of being weak at "proper nouns"I thought that this was something that wasn't widely known.

 

For example, suppose you are recording a meeting and using speech recognition to turn it into minutes. The conversation contains the following proper nouns:

・Company name: "Wow! Lemur!" Co., Ltd.

・Product name: "Wao!Wao!Voice!!"

・Personal name: “Fox Sarutaro”

 

In this example, "Wow! Lemur, Inc." is an original word I came up with, including the exclamation mark, so it will not be accurately transcribed by voice recognition. In the first place, the above proper nouns are only used by me. When using voice recognition for another conversation, it would be problematic if exclamations were added automatically to words like "wow" and "lemur," right?

 With voice recognition, everyone will probably use it to some extent.The "word candidates" are narrowed down, and only pre-learned words are returned.It is like that.

 

 If you register words, it can also recognize proper nouns!

What on earth?!

 

(AmiVoice Cloud Platform) has a word registration function.Apparently, if I register the "proper nouns" that only I use, I might be able to create a voice recognition system just for me.

That's amazing!

 

So let's try it out right away!!

You can register words from your My Page on the AmiVoice Cloud Platform.

 

Then, I registered "Wow! Lemur Co., Ltd.", "Wao!Wao!Voice!!", and "Fox Monkey Taro" as shown below.

 

Let's check if it shows up as a recognized result!

 

Yay, we found a proper noun!

Now you can create your own voice recognition!

 

Mistakes: Be careful of this!! Profile ID

There's one thing I'm stuck on while writing this article: profile ID.

After registering a word, you need to enter your profile ID in order for it to be recognized. It took me about an hour to realize this.

After registering a word, write your profile ID.Please don't forget.

 

20220105 Update

The profile ID field in the javascript sample has been commented out in the update.

 

To display the profile ID, uncomment it.

This is where you should delete it.

 

Bonus: Here's what's great about AmiVoice

■ Word registration function

There will be proper nouns that only you use.

This article explains how to use it.

■ The recognition result does not include the word "Iyaaan"

AmiVoice does not produce unpleasant words (words with sexual or discriminatory meanings) in its recognition results.
If a TV subtitle system were to produce a "yikes" word, it would undoubtedly send chills through the living room. AmiVoice has taken measures to prevent such "yikes" from appearing.

*However, this only applies to words that Advanced Media deems inappropriate.

■ Area-specific engine

The vocabulary used varies depending on the industry.

In banking, you don't hear the medical term "traumatic shoulder dislocation," so the vocabulary used in the banking and medical industries is different.

AmiVoice offers separate voice recognition engines for each industry, offering them as "area-specialized engines."If you have decided on an industry, the recognition rate is much higher, so give it a try.

 

AmiVoice's speech recognition technology has been used in the business field for over 20 years. Over that time, it has grown into a speech recognition technology specialized for business.

 

Finally

This article was about the word registration function, which is rare in other speech recognition APIs.

If you are interested, I would be happy if you would use it.

 

Acknowledgments

I would like to thank "AiAi Senpai" and "Meganezaru Senpai" for their advice and guidance on how to use the API in order to write this article.