Voice Commands Widget

Why use BNF format?

Voice command systems rely on grammars to interpret spoken input accurately. BNF lets you define the precise structure of valid commands, ensuring the system only recognizes intended phrases and rejects ambiguous or invalid inputs.

Here is a brief example of a voice commands grammar file. For more details, please refer to our dedicated documentation.

#BNF+EM V2.1;

!grammar VoicePickGrammar;

// Start rule for the grammar
!start <main>;

// Main rule that encompasses various ways someone might initiate a voice pick
<main>: <direct_pick> [<item>] [<quantity>];

// Direct commands for picking an item
<direct_pick>: "pick" | "select";

// Optional slot for specifying the item to pick
<item>: "item" | "product" | "object" | "thing";

// Optional slot for specifying the quantity of items to pick
<quantity>: "one" | "two" | "three" | "four" | "five";

...

How to use the Grammar AI Generator ?

Elaborate your prompt and paste it (1) to describe the expected behavior of your voice commands. The AI respects BNF standards and provides comments on its design choices.

Click Generate (2) to obtain a ready-to-use grammar in under 5 seconds. There are no limits, aside from decency, on using our AI tools within your project licensing.

Use Create a New Model (3) to save your current grammar. Your project can store an unlimited number of models.

Use Replace current grammar (4) to apply your newly generated grammar directly. Please note that lost grammar cannot be restored.

How to use the Translation AI Tool ?

Choose the Target Language (1), knowing that the source language is inherited from the Model’s Lang (2). Click Translate (3) to generate ready-to-use grammar in under 5 seconds. There are no limits, aside from decency, on using our AI tools under your project licensing. Finish by selecting Create as New Model (4).

Slots are added directly in the grammar editor using the !slot rule. For example:

[...]
!slot <cities>;
<main>: "yes" | "no" | <cities>;

By doing so, the Dynamic Slot Values tab will show a red pastille showing a slot is detected.

Define Slot Values manually

After saving the model, the slot is created and you can edit it using the Edit Slot button (1). It is also possible to edit slot values using a .csv file via the Bulk Import / Export Slots button (2).

Press the plus icon button (1) to add an editable Slot Value. Edit this value using the dedicated fields (2). Press the delete icon button (3) to remove the unecessary values and finally press Confirm button (4) to save.

Define Slot Values from CSV file

If you pressed the Bulk Update Slots button in the Dynamic Slot Value tab, you should see this modal appear.

You can press the Download template button (1a) to obtain the .csv file template or directly download the actual slots in the Unit Test you are editing using the Download Current Slots button (1b).

CSV Structure

The CSV file must contain two required columns (in any order):

• slot_name: Name of the slot

• entry: One value added to the slot

Example CSV (assets.csv):

slot_name,entry
city,Paris
city,London
city,New York
color,red
color,blue
color,green

Import Behavior

When you import this CSV (or a modified version):

1. CSV is parsed with flexible column ordering

2. New slots are ignored if not existing (because you must create them using the grammar !slot rule)

3. Existing slot values are removed to save the .csv file values

4. Existing slots are untouched if not present in the .csv file

5. Invalid rows are ignored

The names are stricly compared during import.

city != cities

color is missing

Rename a slot

The slots must be defined in the grammar of the model. So in order to rename a slot, you have to edit it directly in the editor. For example:

[...]
!slot <city>;
!slot <color>;
<main>: "yes" | "no" | <city> | <color>;

The grammar parser detects we changed the slot, but doesn’t know we are renaming it.

Cities slot values will be lost if we save now.

We can transfer the values of the slot <cities> by clicking the <city> slot in the choices.

Previous Slot values are erased by the process, because VDK Studio consider you exported Slot values before editing them and importing them back.

In the example New York was in the file, not Madrid.

How to read hypothesis ?

The key information is the Start rule. Most grammars have a single main rule, though multiple starting rules can be useful in certain cases.

• A confidence score of over 6000 (above the 4000 threshold) allows the hypothesis to be displayed.

• Alternative hypotheses contain the same information as the main hypothesis.

The hypothesis details show the confidence score for each entity within your voice commands, inside the whole utterance. It allows you to understand which part of the voice command is less understood by the technology.

What are recognizer options ?

Based on this documentation page containing Recognizer parameters, we provide the most used parameters that have a real impact in Quick Testing. Remember you can still change those parameters after Downloading the project.

TSILENCE: The silence period required to detect the Voice Command has elapsed. Any voice recognized afterward will be treated as a new Voice Command.

STREAM_RESULT_MODE: This means results will appear while you speak. Even if you do not pass the TSILENCE, "partial" results will still appear.

MAXNBEST: This value represents the maximum number of hypotheses considered and displayed in the test.