SpeechSmith AI

1. Configure API Keys

SpeechSmith AI uses external AI providers for text analysis. You'll need to configure at least one of the following:

• OpenAI API Key (supports GPT-4 and GPT-3.5 models)
• Anthropic Claude API Key (supports Claude 3 models)

Configure these keys in your Dashboard AI Settings.

2. Access SpeechSmith AI

Navigate to SpeechSmith AI from the Tools section or directly via https://cxtools.xyz/tools/speechsmith

3. Add Your IVR Script

Paste your IVR script in the input area. Your script should include the prompts and expected user responses.

4. Configure Settings

Adjust your SpeechSmith settings to customize the GRXML generation:

• Language: Select the primary language for your IVR (e.g., en-US, es-ES)
• Industry: Select your industry for context-specific optimizations
• Confidence Threshold: Set the minimum confidence level for phrase matching
• Variations: Choose which types of variations to generate (plural forms, polite variations, regional variations)
• AI Model: Select your preferred AI model (based on configured API keys)

5. Generate GRXML

Click the "Analyze Script" button to process your script and generate GRXML. SpeechSmith will:

• Extract key phrases from your script
• Generate variations for each phrase
• Create a comprehensive GRXML file with proper structure
• Add CXone Studio-compatible tags and attributes

6. Review and Edit

After generation, you'll see:

• The complete GRXML code
• A list of extracted phrases with alternatives
• Confidence scores for each phrase

Review the generated content and make any necessary adjustments.

Alternative Suggestion Generator

For any phrase in your project, you can generate additional alternatives:

1. Select a phrase from your project or enter a new one
2. Click "Generate Alternatives"
3. Review the list of AI-generated alternatives
4. Add selected alternatives to your phrase list

GRXML Testing Tool

Test how well your GRXML recognizes specific phrases:

1. Enter a test phrase
2. Click "Test Phrase"
3. Review match results, including:
• Whether the phrase matched
• The specific grammar item that matched
• The confidence score for the match
• Explanation of why it matched or didn't match

This feature helps you identify and fix potential recognition issues before deploying your IVR.

Project Management

Save and manage your SpeechSmith projects for future reference or iteration:

• Save Project: Give your project a name and description
• Load Project: Access previously saved projects
• Export GRXML: Download the generated GRXML file for use in CXone Studio
• Delete Project: Remove projects you no longer need

Provide Comprehensive Scripts

Include as much context as possible in your IVR scripts. The more information SpeechSmith has, the better it can generate appropriate phrases and variations.

Test with Real-World Phrases

Use the testing tool with phrases that real customers might use, including:

• Informal language and slang
• Incomplete or fragmented sentences
• Regional variations and accents
• Different ways of expressing the same intent

Incorporate Customer Data

If available, use actual customer utterances from your existing IVR data to test and refine your GRXML. This helps ensure your grammar matches real-world usage patterns.

Regularly Update Your Grammars

Language and customer expectations evolve. Regularly review and update your GRXML to incorporate new phrases and patterns that may emerge over time.

API Key Issues

If you encounter API key errors:

• Verify your API key is correctly entered in your dashboard settings
• Check that your API key has sufficient credits or quota
• Try a different AI provider if available
• Ensure your API key has the correct permissions

Generation Failures

If GRXML generation fails:

• Try again with a simpler script or break your script into smaller sections
• Check for unusual characters or formatting in your script
• Try a different AI model that may handle your specific use case better
• Ensure your script is clear and focused on specific intents

Poor Recognition Results

If your GRXML isn't recognizing phrases well:

• Add more explicit variations for commonly used phrases
• Test with different phrasings to identify gaps in your grammar
• Adjust your confidence threshold settings
• Consider industry-specific terminology that may need special handling
• Check for overlapping phrases that may cause confusion

Which AI models work best for GRXML generation?

In our testing, both Claude and GPT-4 models perform excellently for GRXML generation. Claude models tend to generate more structured GRXML output, while GPT-4 may generate more creative phrase variations. For most use cases, we recommend using the Claude 3.5 Sonnet model for the best balance of performance and cost.

Can SpeechSmith handle languages other than English?

Yes, SpeechSmith supports multiple languages. When generating GRXML, select the appropriate language in the settings panel. The tool currently supports English, Spanish, French, German, Italian, Portuguese, and Japanese. Performance may vary by language depending on the AI model used.

How much does it cost to use SpeechSmith AI?

SpeechSmith AI is included in the CX Tools Pro subscription. Additionally, you'll need your own API key for OpenAI or Anthropic. The API usage costs depend on the models you use and the volume of script processing. A typical GRXML generation for a medium-sized IVR script costs approximately $0.10-0.30 in API fees.

Is my data secure?

CX Tools does not store the content of your scripts or generated GRXML on our servers beyond the current session, unless you explicitly save it as a project. All processing is done using secure connections to the AI providers. If you have specific data security requirements, you should review the privacy policies of the AI provider you choose (OpenAI or Anthropic).

Can SpeechSmith generate grammars from existing NICE CXone scripts?

Yes, SpeechSmith can generate GRXML from text extracted from existing CXone scripts. Copy the relevant prompt text and expected responses from your script and paste them into SpeechSmith. For best results, format the input to clearly indicate what the system says versus what users might respond with.