Skip to content

Latest commit

 

History

History
82 lines (63 loc) · 3.84 KB

using-grammars-for-speech-recognition.md

File metadata and controls

82 lines (63 loc) · 3.84 KB
Raw

Using Grammars for Speech Recognition

This guide is to show you how to use grammars with IBM Voice Gateway. With the IBM Speech-To-Text service one can uses grammars to specify how to form valid strings from the language's alphabet. When you use a custom language model and a grammar for speech recognition, the service can return a valid phrase from the grammar or an empty result. Grammars are useful when you need to recognize specific words or phrases, such as yes or no, individual letters, numbers as well as alphanumerical combinations.

For more information on grammars with IBM Speech-To-Text service https://cloud.ibm.com/docs/services/speech-to-text?topic=speech-to-text-grammars#grammars

First you'll need to create a custom language model as well as a grammar to use, you can follow the steps defined in: Adding a grammar to a custom language model

Here are a some options of using grammars with IBM Voice Gateway:- Using Grammars for Speech Recognition

Setting a grammar from Watson Assistant

Generally, you may want to use a grammar to recognize some kind of an expected phrase, for example matching an alpha numerical combination. By using the vgwActSetSTTConfig action from a Watson Assistant Dialog node you can modify the Speech recognition configuration to use grammar for just an input from the user or you can set it for the rest of the conversation.

For example, in your Watson Assistant Dialog node, specify the vgwActSetSTTConfig, as the action vgwAction to use, it would use the following structure:

{
  "output": {
    "text": {
      "values": [
        "Hey, testing a grammar input!"
      ],
      "selection_policy": "sequential"
    },
    "vgwAction": {
      "command": "vgwActSetSTTConfig",
      "parameters": {
        "updateMethod": "mergeOnce",
        "config": {
          "grammar_name": "{your_grammar_name}",
          "language_customization_id": "{your_language_customization_id}"
        }
      }
    }
}

As you can see, updateMethod set to mergeOnce specifies that the grammar will be merged and used once on the user input. If it is set to merge, the grammar will be merged to the existing speech recognition configuration for the rest of the conversation.

For more information on the vgwActSetSTTConfig action, see Dynamically configuring the Speech to Text service for more information.

Advanced JSON Configuration

You can specify it through the Advanced JSON Configuration. This configuration will be used for all recognition requests by the IBM Voice Gateway. You can modify the configuration using the vgwActSetSTTConfig action from a Watson Assistant Dialog node, see Dynamically configuring the Speech to Text service for more information.

{
  "tenants": [{
    "tenantURI": "2345556789",
    "description": "Voice Gateway Demo",
    "conversation": {
        ...
    },
    "stt": {
      "credentials": {
        "url": "https://stream.watsonplatform.net/speech-to-text/api",
        "username": "9h7f54cb-f28f-4a64-91e1-a0657e1dd3f4",
        "password": "password"
      },
      "config": {
        "model": "en-US_NarrowbandModel",
        "language_customization_id": "{your_customization_id}",
        "grammar_name": "{your_grammar_name}"
        "profanity_filter": true,
        "smart_formatting": true
      }
    },
    "tts": {
        ...
    }
  }]
}