Transcripts

Every call in ScaleTalk generates a real-time transcript, capturing both the caller's and agent's speech with accurate timestamps.

Transcript Overview

Transcripts provide:

• Full Conversation Text - Complete dialogue
• Speaker Identification - Caller vs. Agent labels
• Timestamps - When each utterance occurred
• Search Capability - Find specific content

Viewing Transcripts

From Call Detail

1. Navigate to Calls
2. Click on a specific call
3. Scroll to the Transcript section

Transcript Format

[00:00:02] Agent: Hello, this is ScaleTalk support. How can I help you today?

[00:00:08] Caller: Hi, I'd like to schedule an appointment for next week.

[00:00:12] Agent: I'd be happy to help you with that. What day works best for you?

[00:00:18] Caller: How about Tuesday morning?

Speaker Identification

Labels

Accuracy

Speaker detection uses:

• Audio channel separation
• Voice activity detection
• Pattern recognition

Searching Transcripts

Within a Call

1. Open the transcript view
2. Use Ctrl/Cmd + F to search
3. Navigate between matches

Across All Calls

1. Go to the Calls page
2. Use the search bar
3. Enter keywords or phrases
4. Filter by date range

Search Tips

• Use quotes for exact phrases: "schedule appointment"
• Search for caller names mentioned
• Look for specific products or topics

Variable Extraction

ScaleTalk can automatically extract structured data from transcripts.

Configured Variables

Define what to extract in your agent's Follow-Up settings:

How Extraction Works

1. Call ends
2. Transcript is analyzed
3. LLM extracts specified variables
4. Data appears in call details
5. Webhooks send to external systems

Configuration

In the agent's Follow-Up tab:

Variables to Extract:
- caller_name: The caller's full name
- phone_number: Callback number if provided
- issue_type: Category of the inquiry
- resolved: Whether the issue was resolved (yes/no)

Export Options

Single Transcript

1. Open call details
2. Click Export > Transcript
3. Choose format (TXT, JSON)

Bulk Export

1. Go to Calls
2. Apply filters
3. Click Export > Transcripts
4. Download as CSV or JSON

Export Formats

Text Format (TXT):

Call ID: call_123456
Date: 2024-01-15 10:30:00
Duration: 4:05

[00:00:02] Agent: Hello...
[00:00:08] Caller: Hi...

JSON Format:

{
  "call_id": "call_123456",
  "transcript": [
    {
      "speaker": "agent",
      "text": "Hello...",
      "timestamp": 2.0
    },
    {
      "speaker": "caller",
      "text": "Hi...",
      "timestamp": 8.0
    }
  ]
}

Accuracy and Quality

Speech Recognition

ScaleTalk uses Deepgram Nova-3 for transcription:

• Real-time streaming
• High accuracy for German and English
• Custom vocabulary support

Improving Accuracy

For better transcription:

1. Ensure good audio quality
2. Train custom vocabulary in settings
3. Use clear, professional voice models

Known Limitations

• Background noise may affect accuracy
• Strong accents may require tuning
• Technical terms may need vocabulary training

Privacy Considerations

Data Handling

Transcripts are:

• Encrypted at rest
• Access-controlled
• Subject to retention policies

Redaction (Enterprise)

Enterprise plans support:

• PII detection
• Automatic redaction
• Manual review options

Integration with Webhooks

Send transcripts to external systems:

{
  "event": "call.completed",
  "data": {
    "call_id": "call_123456",
    "transcript": "...",
    "variables": {
      "caller_name": "John Smith"
    }
  }
}

See Follow-Up for configuration details.