Rss Categories

Pronunciation Checker

Reference Number: AA-01953 Views: 2568 0 Rating/ Voters

The Pronunciation Checker displays the list of words in a selected grammar and indicates how the ASR determines the pronunciation of the word. This can be used to find words which may benefit from custom lexicon entries in order to help with recognition issues. The Pronunciation Checker was introduced in the 12.2 version of the Speech Tuner and requires a 12.2 or newer version of the ASR to function.

Accessing the Pronunciation Checker

To access the Pronunciation Checker, load a grammar into the Speech Tuner. To access the Pronunciation Checker for the current grammar in the Grammar Editor, simply go to Tools > Pronunciation Checker:

Or you can right-click any grammar in the Grammar List and choose Pronunciation Checker:

Understanding Pronunciation

For each word in the grammar, the Pronunciation checker displays a color-coded symbol to the left of the word indicating where its pronunciation comes from:

  • Green indicates the word is in the built-in main dictionary used by LumenVox for the current language.
  • Yellow indicates that the pronunciation is not in a dictionary or lexicon, and is instead generated by the LumenVox ASR's rules.

To get the most value out of the Pronunciation Checker, it is helpful to understand how exactly the LumenVox ASR determines how words are pronounced and what that means. Pronunciation is how a word sounds, and the ASR needs to know the various ways a word can be pronounced in order to best convert audio into text. The most common way it does this is by relying on its built-in dictionary, which contains tens or hundreds of thousands of words and their phonetic pronunciation.

Given that there exist millions of words, however, it is not realistic that every single word can be included in the built-in dictionary. If a word is not in the dictionary, the ASR relies on both built-in rules about how words are pronounced in a language as well as a complex statistical model that gives it an idea of the probability that a given combination of letters is pronounced a certain way. Generally speaking, these statistical and rules-based methods produce good results.

For cases where those methods don't work, custom lexicons allow developers to supply their own pronunciations. These are most useful for items like uncommon names or foreign words that may be pronounced differently than most words in a language.

The Pronunciation Checker allows developers to check, at a glance, and see if any words in their grammar might benefit from adding custom lexicons. Words which are yellow (not in a dictionary or lexicon) are good candidates for lexicon entries, particularly if you notice frequent misrecognitions of those words (the Accuracy Tuning Wizard may help you identify such words).

To learn how the ASR believes a word is pronounced, you can select it in the list of words and then click Phonetic Spelling... to see the word spelled phonetically by the Phonetic Speller. You can use this to help identify words where the pronunciation generation rules may not be working optimally.