Automatic Speech Recognition

Compatibility of ASR module output format and NLU module input format

Depending on the ASR module used to convert speech to text, there are different rules for the spelling of training phrases in voice bots so that their format is consistent with the format appropriate for both the NLU engine and the output from the ASR module.

For example, no punctuation marks (commas, periods), no capital letters or numerals written in some cases with text (with words) and writing without typos, etc. The full set of rules will be provided by SentiOne depending on the selected ASR module.

A similar set of spelling rules is given for bot and speech synthesizer responses. In both cases, you can report phonetic exceptions or recognition for new words or proper names and test such recognition or pronunciation in the demo.

TIPS

It is worth verifying the correctness of phrases in the set from time to time (e.g. once every 2 weeks), especially if more than one person is working on it,
To balance the set, it is worth including phrases with both feminine and masculine endings for verbs for applicable languages (like German, Spanish or Polish), e.g. (I would like to/activated/activated), adverbs (alone/alone/alone) and adjectives (I am interested/interested) about the author / sender as well as its recipient (can you check it out for me), but you do not need to add the same phrase twice if the only difference is the gender of the interlocutor.
*language dependent

Defining ASR model in bot flow

You can control which ASR model can be used in bot by passing the recognizeModel command with one of the following values:

spelling - literal spelling of words
name - First names as well as surnames
address - Locations, addresses, names of cities etc.

Commands can be added in block when you request data from user. Custom model would be used only in this specific block.

Please note that availability of the command depends on the configuration of the Voice Integrator and may not be available in every project.