Skip to main content

Tips

Number of registered words​

You may register around 1,000 words per profile. You can only register up to 1,000 words from MyPage.

Each engine has a predetermined vocabulary, and words not included in that vocabulary will never be recognized. Even if you exceed 1,000 words, it may still be beneficial to register them. However, be aware that newly added words may be mistakenly recognized as other words with similar pronunciations. Adding too many words can increase such errors and potentially decrease recognition accuracy.

Additionally, increasing the number of registered words has drawbacks such as longer loading times for profiles during speech recognition and increased speech recognition time. Particularly, the profile loading time increases exponentially with the number of registered words. Similarly, use caution when specifying words in the profileWords request parameter during requests.

Guidelines​

As general precautions, please keep the following in mind:

  • When a corresponding class is available, specify the class when registering words.
  • Words with few syllables are more likely to be misrecognized, so either avoid adding them or always specify a class.
  • Don't register words with excessively long pronunciations. If you want to register proper nouns or words with long pronunciations, consider splitting the word and registering it in parts. For example, if you want to register the station name "南阿蘇水の生まれる里白水高原駅", one approach is to register "南阿蘇" and "白水高原" in the '地名' class, which is used for Japanese place names, excluding the general words that can likely be recognized without registration.
  • Don't register words with the same pronunciation in the same class. When registering words with the same pronunciation in different classes, you can expect them to be distinguished based on context, but keep in mind that perfect distinction is not guaranteed.
  • Don't use word registration to rewrite recognition results to expected notations. For instance, in cases where you want "子ども" to be recognized but the recognition result outputs "子供". It's not recommended to register words in an attempt to get the desired recognition result in such cases. Registered words are not always recognized, and there's a risk of misrecognition of unspoken registered words. If you want to rewrite recognition results to expected notations, please perform post-processing in the client program.
  • Don't add words that are unlikely to be spoken just as a precaution. If you have a word list for a service or industry, don't register all of them at once. Instead, consider the above points and carefully select which words to register.