Familiarity with Voci products is helpful when developing and deploying hints. Voci online help discusses terminology, data flows, and APIs that provide an invaluable foundation for getting the most out of the hinting feature.

Good hints are words or short phrases that are important to a customer's business that are likely to be used in a call or set of calls and are important to transcribe correctly. Each hint tunes Voci's ASR engine to improve recognition accuracy of a specific word or phrase, so that a more accurate transcript is produced. Hints are typically proper nouns or noun phrases that are specific to a company or organization, but they can be other types of short phrases such as those found in legal disclaimers or mini-Miranda scripts.

An example of a hint is "Joe's Hardware". These two words do not appear next to each other often in typical speech, but they appear next to each other frequently in customer-care calls received on behalf of a store named "Joe's Hardware".

Because those words do not occur together with high frequency in general speech, Voci 's off-the-shelf language models consider such a word pairing to be unlikely. Without hinting, this phrase will be transcribed at lower accuracy than the rest of the conversation.

If you provide the hint "Joe's Hardware" for a set of calls, the ASR engine is informed that "Joe's Hardware" is a word pairing that is common in that set of calls. Then, if the ASR engine finds multiple possible interpretations of speech in that set of calls, it is more likely to select the hinting phrase if that hinting phrase is one of the possible interpretations. The practical effect is that the phrase "Joe's Hardware" is more likely to be accurately recognized when it is spoken than it would be without the provided hint.

A hint can include words that are not in the dictionary of the language model being used. In other words, it is acceptable for hints to include out of vocabulary (OOV) words. The Voci ASR system will deduce the North American English pronunciation of OOV words in hints, and the recognition of those words is improved. Examples include hints like "Tilex" and "Briggs and Stratton".

Hints are organized into categories such as company_name and brand_name. This helps the ASR engine understand the context likely to surround hints. For example, a company name is likely to appear after the phrase "Thank you for calling", but a brand name is not. Likewise, a brand name is likely to appear after the phrase "I bought a" or "How much is a", whereas a company name is not.

This is similar to how spelling bee contestants ask to hear an unfamiliar word used in a sentence. The nature of a word and how it fits into surrounding context are highly correlated.

The following sections explain how to develop and deploy hints.