​

Advanced Features

​

Overview

• Language Switching: Let users switch languages mid-conversation
• Ambient Noise Handling: Add realistic background noise to agent voice output
• Call Transfer: Route calls to human agents or other endpoints
• Post-Call Analysis: Extract structured insights from conversations
• Talk First: Agent speaks first when call connects
• Max Duration: Set maximum call duration limits
• IVR Detection: Detect answering machines and IVR systems

​

Language Switching

​

What is Language Switching?

• Respond to user language requests
• Adapt voice synthesis to match the requested language
• Continue the conversation in the target language
• Switch back to the original language when requested

​

When to Use Language Switching

• Customer support for global audiences
• Healthcare services with diverse patient populations
• Government services in multilingual regions
• Travel and hospitality applications

User (in English): "Hello, I need help with my account"
Agent: "Hello! I'd be happy to help. How can I assist you today?"
User: "Sorry, no English. Spanish?"
Agent (in Spanish): "¡Por supuesto! ¿En qué puedo ayudarte hoy?"

​

Configuration

1. Navigate to the Features tab in the agent editor
2. Toggle Enable Language Switching to ON
3. Save your agent configuration

const agent = await fetch('https://blackbox.dasha.ai/api/v1/agents', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    name: "Multilingual Support Agent",
    config: {
      version: "v1",
      primaryLanguage: "en-US",
      features: {
        version: "v1",
        languageSwitching: {
          version: "v1",
          isEnabled: true
        }
      },
      llmConfig: {
        version: "v1",
        vendor: "openai",
        model: "gpt-4o-mini",
        prompt: "You are a multilingual support agent. When users request a language change, respond in their preferred language."
      },
      ttsConfig: {
        version: "v1",
        vendor: "ElevenLabs",
        model: "eleven_multilingual_v2",
        voiceId: "YOUR_VOICE_ID"
      }
    }
  })
});

​

Requirements

1. LLM Support: Use models that handle multiple languages well
   • ✅ OpenAI GPT-4/GPT-4.1 (excellent multilingual support)
   • ✅ DeepSeek (strong multilingual capabilities)
   • ⚠️ Some specialized models may have limited language support
2. TTS Support: Choose voices that match your target languages
   • ElevenLabs Multilingual V2: Supports 29 languages
   • Cartesia Sonic: Multilingual capabilities
   • Dasha: Supports just 2 languages, but fast
3. System Prompt: Instruct your agent to handle language requests
   Copy

   Ask AI

   You are a multilingual customer support agent. When users request
   a language change, continue the conversation in their chosen language.
   Maintain context and conversation flow.
   

​

Ambient Noise Handling

​

What is Ambient Noise?

• Makes the agent sound more realistic and human-like
• Simulates natural environments (office, call center, etc.)
• Adds authenticity to AI-generated voice
• Creates a more natural conversation experience

​

When to Use Ambient Noise

• Agents posing as human customer service representatives
• Scenarios where perfect audio quality seems artificial
• Call center simulations requiring realistic background
• Applications where environmental context matters

• Customer service agent appearing to work from an office
• Sales representative calling from a realistic workplace
• Appointment reminders that sound naturally human
• Any scenario where pure AI voice seems too perfect

​

Configuration

1. Navigate to the Features tab
2. Toggle Ambient Noise to ON
3. Adjust the noise level from 0.0 (none) to 1.0 (maximum)
4. Test to find the right balance for your use case

const agent = await fetch('https://blackbox.dasha.ai/api/v1/agents', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    name: "Realistic Voice Agent",
    config: {
      version: "v1",
      primaryLanguage: "en-US",
      features: {
        version: "v1",
        ambientNoise: {
          version: "v1",
          isEnabled: true,
          ambientNoiseLevel: 0.4  // 0.0 to 1.0 range
        }
      },
      llmConfig: {
        version: "v1",
        vendor: "openai",
        model: "gpt-4o-mini"
      },
      ttsConfig: {
        version: "v1",
        vendor: "ElevenLabs",
        voiceId: "YOUR_VOICE_ID"
      }
    }
  })
});

​

Finding the Right Level

• 0.0: No background noise - pure AI voice (most artificial)
• 0.2-0.4: Subtle office ambiance - natural without distraction
• 0.5-0.7: Moderate background - noticeable call center environment
• 0.8-1.0: Heavy ambient noise - may impact clarity

​

Best Practices

• Use enough noise to sound natural
• Avoid levels that impair understanding
• Test with your target audience

Office environment: 0.2-0.4
Call center: 0.4-0.6
Natural outdoor setting: 0.5-0.7

• Some jurisdictions require AI disclosure
• Ambient noise doesn’t replace legal requirements
• Be transparent about AI agent identity when required

​

Call Transfer

​

What is Call Transfer?

• Escalate complex issues to human agents
• Route calls to departments or specialists
• Forward calls to external phone numbers
• Maintain conversation context during transfer

​

Transfer Types

​

Cold Transfer (Blind Transfer)

• Simple routing scenarios
• No context needed at destination
• Fast handoff required
• IVR-style call routing

Agent: "Let me transfer you to our billing department."
[Agent initiates transfer and disconnects]
[User connects directly to billing]

​

Warm Transfer (Attended Transfer)

• Complex issues requiring context
• VIP or sensitive situations
• Quality assurance needed
• Human agent availability check

Agent: "I'm going to connect you with a specialist. Please hold."
[Agent calls specialist privately]
Agent (to specialist): "I have a customer with billing issue X..."
Specialist: "I can help. Transfer them over."
[Agent completes transfer]

​

HTTP Transfer (Programmatic Routing)

• Business rule-based routing
• CRM integration for routing
• Load balancing across agents
• Custom transfer logic

// Your webhook receives transfer request
{
  "callId": "call_123",
  "agentId": "agent_456",
  "reason": "billing_issue",
  "context": { "accountId": "acc_789" }
}

// Return transfer destination
{
  "isSuccess": true,
  "transfer": {
    "version": "v1",
    "type": "warm",
    "isEnabled": true,
    "endpointDestination": "+1-555-0199", // or SIP URI
    "description": "Transfer to specialist for account issues"
  },
  "message": "Transfer initiated successfully"
}

​

Configuration

1. Navigate to the Features tab
2. Toggle Call Transfer to ON
3. Select transfer type (cold, warm, or HTTP)
4. Configure destination endpoint(s)
5. Add transfer instructions to system prompt

{
  config: {
    version: "v1",
    features: {
      version: "v1",
      transfer: {
        version: "v1",
        type: "cold",
        isEnabled: true,
        endpointDestination: "+1-555-0100",
        description: "Transfer to billing department for payment questions"
      }
    }
  }
}

{
  config: {
    version: "v1",
    features: {
      version: "v1",
      transfer: {
        version: "v1",
        type: "warm",
        isEnabled: true,
        endpointDestination: "sip:specialist@example.com",
        description: "Warm transfer with full context briefing",
        interactionWithOperator: {
          type: "static"
        }
      }
    }
  }
}

{
  config: {
    version: "v1",
    features: {
      version: "v1",
      transfer: {
        version: "v1",
        type: "http",
        isEnabled: true,
        description: "Dynamic routing via webhook",
        webhook: {
          url: "https://your-api.com/transfer-routing",
          headers: {
            "X-Custom-Header": "value"
          }
        },
        fallback: {
          version: "v1",
          type: "cold",
          isEnabled: true,
          endpointDestination: "+1-555-0100",
          description: "Fallback to main support"
        }
      }
    }
  }
}

app.post('/transfer-routing', async (req, res) => {
  // Request payload structure (TransferWebHookPayload)
  const {
    callId,
    agentId,
    orgId,
    callAdditionalData,
    agentAdditionalData,
    transferReason,
    transcription,
    status,  // Always "Running"
    type     // Always "TransferWebHookPayload"
  } = req.body;

  // Your routing logic based on call context
  const destination = await determineDestination(callAdditionalData);

  // Response must follow TransferWebhookResponse schema
  res.json({
    isSuccess: true,
    transfer: {
      version: "v1",
      type: "cold",
      isEnabled: true,
      endpointDestination: destination,
      description: "Routed based on call context"
    },
    message: "Transfer initiated successfully"
  });

  // For errors, return isSuccess: false
  // res.json({
  //   isSuccess: false,
  //   message: "Unable to determine destination"
  // });
});

​

Transfer Routes

{
  config: {
    version: "v1",
    features: {
      version: "v1",
      transfer: {
        version: "v1",
        type: "cold",
        isEnabled: true,
        endpointDestination: "+1-555-0100",
        description: "Main routing agent with multiple destinations",
        transferRoutes: [
          {
            name: "billing",
            endpointDestinations: [
              { endpoint: "+1-555-0100" }
            ],
            description: "Billing department for payment questions"
          },
          {
            name: "technical",
            endpointDestinations: [
              { endpoint: "sip:tech@example.com" }
            ],
            description: "Technical support for product issues"
          },
          {
            name: "emergency",
            endpointDestinations: [
              { endpoint: "+1-555-0911" }
            ],
            description: "Urgent issues requiring immediate attention"
          }
        ]
      }
    }
  }
}

You are a customer service routing agent. Route calls as follows:

- Billing questions → transfer to billing route
- Technical problems → transfer to technical route
- Emergencies → transfer to emergency route immediately
- General questions → handle directly

Always explain to the user why you're transferring them and
what to expect.

​

Transfer Best Practices

Agent: "I'm going to connect you with our billing specialist
who can help with your payment plan. This should take just
a moment."

Agent (to specialist): "I have Jane Doe on the line with a
question about invoice #12345. She's asking about payment
terms."

{
  config: {
    version: "v1",
    features: {
      version: "v1",
      transfer: {
        version: "v1",
        type: "http",
        isEnabled: true,
        description: "HTTP transfer with fallback",
        webhook: {
          url: "https://your-api.com/routing"
        },
        fallback: {
          version: "v1",
          type: "cold",
          isEnabled: true,
          endpointDestination: "+1-555-0101",
          description: "Backup destination if webhook fails"
        }
      }
    }
  }
}

• Transfer success rate
• Time to connect
• Post-transfer customer satisfaction
• Reasons for transfer

​

Testing Transfers

• Enable transfers
• Configure test phone numbers
• Set short timeout values

1. Successful cold transfer
2. Successful warm transfer
3. Transfer failure (busy/no answer)
4. Multiple transfer attempts
5. Fallback routing

• Call connects to destination
• Context is preserved (warm transfer)
• Fallbacks work correctly
• User experience is smooth

​

Post-Call Analysis

​

What is Post-Call Analysis?

• Extract key information from conversations
• Categorize call outcomes
• Score conversation quality
• Generate structured data for your systems

​

Use Cases

• Sentiment analysis
• Issue categorization
• Resolution status
• Follow-up requirements

• Lead qualification scores
• Interest level assessment
• Next steps identification
• Objection tracking

• Symptom documentation
• Appointment scheduling confirmation
• Insurance information capture
• Follow-up instructions

​

Configuration

const agent = await fetch('https://blackbox.dasha.ai/api/v1/agents', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    name: "Sales Qualification Agent",
    config: {
      version: "v1",
      features: {
        version: "v1",
        postCallAnalysis: [
          {
            version: "v1",
            name: "lead_qualification",
            isEnabled: true,
            labels: [
              {
                name: "lead_score",
                type: "number",
                description: "Interest level from 1-10"
              },
              {
                name: "budget_confirmed",
                type: "boolean",
                description: "Did prospect confirm budget availability?"
              },
              {
                name: "timeline",
                type: "enum",
                values: ["immediate", "1-3_months", "3-6_months", "6plus_months", "unknown"],
                description: "Purchase timeline"
              },
              {
                name: "main_pain_point",
                type: "string",
                description: "Primary challenge the prospect mentioned"
              }
            ]
          }
        ]
      }
    }
  })
});

​

Label Types

{
  name: "customer_concern",
  type: "string",
  description: "Main issue raised by customer"
}

{
  name: "issue_resolved",
  type: "boolean",
  description: "Was the customer's issue resolved?"
}

{
  name: "satisfaction_score",
  type: "number",
  description: "Customer satisfaction from 1-10"
}

{
  name: "call_outcome",
  type: "enum",
  values: ["resolved", "escalated", "callback_needed", "no_action"],
  description: "Final call outcome"
}

​

Accessing Analysis Results

1. Navigate to Calls page
2. Click on completed call
3. View “Post-Call Analysis” section
4. See extracted labels and values

// Post-call analysis is available in the call result object
const response = await fetch(
  'https://blackbox.dasha.ai/api/v1/callresults/search',
  {
    method: 'POST',
    headers: {
      'Authorization': 'Bearer YOUR_API_KEY',
      'Content-Type': 'application/json'
    },
    body: JSON.stringify({
      callIds: ['CALL_ID']
    })
  }
).then(r => r.json());

const callResult = response.results[0];
const analysis = callResult.result?.postCallAnalysis;
// {
//   lead_qualification: {
//     lead_score: 8,
//     budget_confirmed: true,
//     timeline: "1-3_months",
//     main_pain_point: "Manual data entry taking too much time"
//   }
// }

// Completed webhook payload includes result.postCallAnalysis
{
  "type": "CompletedWebHookPayload",
  "status": "Completed",
  "callId": "call_123",
  "agentId": "agent_456",
  "orgId": "org_789",
  "callAdditionalData": {},
  "agentAdditionalData": {},
  "callType": "OutboundAudio",
  "createdTime": "2024-01-15T10:30:00Z",
  "completedTime": "2024-01-15T10:35:30Z",
  "durationSeconds": 330,
  "inspectorUrl": "https://blackbox.dasha.ai/calls/call_123",
  "transcription": [...],
  "result": {
    "postCallAnalysis": {
      "lead_qualification": {
        "lead_score": 8,
        "budget_confirmed": true,
        "timeline": "1-3_months",
        "main_pain_point": "Manual data entry"
      }
    },
    "transferInformation": []
  }
}

​

Combining Features

​

Example: Multilingual Support with Transfer

{
  config: {
    version: "v1",
    primaryLanguage: "en-US",
    features: {
      version: "v1",
      languageSwitching: {
        version: "v1",
        isEnabled: true
      },
      transfer: {
        version: "v1",
        type: "cold",
        isEnabled: true,
        endpointDestination: "+1-555-0100",
        description: "Transfer to language-specific support teams",
        transferRoutes: [
          {
            name: "english_support",
            endpointDestinations: [
              { endpoint: "+1-555-0100" }
            ],
            description: "English support team"
          },
          {
            name: "spanish_support",
            endpointDestinations: [
              { endpoint: "+1-555-0101" }
            ],
            description: "Spanish support team"
          },
          {
            name: "french_support",
            endpointDestinations: [
              { endpoint: "+1-555-0102" }
            ],
            description: "French support team"
          }
        ]
      }
    }
  }
}

You are a multilingual support router. Detect and respond in the user's
language. For complex issues requiring human assistance, transfer to the
appropriate language support team based on the conversation language.

​

Example: Post-Call Analysis for Compliance

{
  config: {
    version: "v1",
    features: {
      version: "v1",
      postCallAnalysis: [
        {
          version: "v1",
          name: "compliance_check",
          isEnabled: true,
          labels: [
            {
              name: "sensitive_data_shared",
              type: "boolean",
              description: "PII or sensitive data discussed"
            },
            {
              name: "escalation_needed",
              type: "boolean",
              description: "Issue requires human follow-up"
            }
          ]
        }
      ]
    }
  }
}

​

Best Practices Summary

​

Language Switching

• Ensure your LLM supports multilingual conversations
• Choose TTS providers with strong multilingual capabilities
• Test language detection and switching in realistic scenarios
• Monitor quality and accuracy across languages

​

Ambient Noise

• Start with level 0.3-0.5 for subtle realism
• Balance natural sound with call clarity
• Test with target audience before deployment
• Match noise level to your use case scenario

​

Call Transfer

• Always configure fallback destinations for HTTP transfers
• Brief recipients during warm transfers for context
• Test all transfer flows thoroughly before production
• Monitor transfer success rates and failure reasons

​

Post-Call Analysis

• Define clear, specific labels with detailed descriptions
• Use enums for categorical data to ensure consistency
• Keep label count manageable (5-10 per form)
• Validate analysis accuracy against actual conversations

​

Troubleshooting

​

Language Switching Not Working

• Verify TTS provider supports target languages
• Check LLM has multilingual capabilities
• Ensure voices are available for all languages
• Review system prompt language instructions

​

Ambient Noise Too High or Too Low

• Reduce level if users complain about audio quality
• Increase level if agent sounds too artificial
• Test different levels with various audiences
• Consider use case requirements (realism vs clarity)

​

Transfer Failures

• Verify destination numbers are correct
• Test reachability of transfer endpoints
• Check SIP configuration if using SIP URIs
• Monitor network connectivity
• Configure fallback routes

​

Inaccurate Post-Call Analysis

• Review and refine label descriptions
• Provide examples in descriptions
• Use enums instead of free-form strings
• Test with various conversation types
• Iterate based on results

​

Next Steps

1. Enable Features: Add features one at a time to your agent
2. Test Thoroughly: Verify each feature works as expected
3. Monitor Performance: Track metrics for enabled features
4. Optimize: Refine configuration based on real-world usage

​

Related Documentation

• Voice & Speech Configuration
• Tools & Functions
• Webhooks & Events
• Call Inspector
• Testing Your Agent

​

API Cross-Refs

• POST /api/v1/agents
• PUT /api/v1/agents/{agentId}
• GET /api/v1/calls/{callId}