> ## Documentation Index
> Fetch the complete documentation index at: https://docs.burki.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Azure Speech STT

> Azure Speech-to-text with broad language support, phrase lists, and Microsoft ecosystem integration

<Card title="☁️ Azure Speech: Enterprise Scale" icon="cloud">
  Microsoft's neural STT service with broad language and regional variant support. Seamless integration with Azure ecosystem and phrase lists for term boosting. Good for organizations already using Microsoft services or requiring broad language support.
</Card>

## Quick Setup

<Steps>
  <Step title="Create Azure Speech Resource">
    1. Follow the [Azure AI Speech resource quickstart](https://learn.microsoft.com/azure/ai-services/speech-service/get-started-speech-to-text)
    2. Create a new **Speech** resource
    3. Select your subscription, resource group, and region
    4. Note your **Key** and **Region** from the resource's Keys and Endpoint page
  </Step>

  <Step title="Configure in Burki">
    1. Go to **AI Configuration** → **STT** tab
    2. Select **Azure** as the provider
    3. Enter your **Subscription Key** and **Region** (e.g., `eastus`, `westus2`)
  </Step>

  <Step title="Choose Model & Language">
    Select your preferred model and language from the dropdowns
  </Step>
</Steps>

<Callout type="info">
  **Free Tier**: Azure offers 5 hours of audio per month free for speech-to-text. See [Azure Pricing](#pricing-overview) for details.
</Callout>

***

## Available Models

<CardGroup cols={3}>
  <Card title="🎯 Standard" icon="bullseye">
    **General Purpose**

    Balanced accuracy and performance for most use cases

    **Languages**: 30+
    **Best for**: General transcription
  </Card>

  <Card title="⚡ Enhanced" icon="bolt">
    **Improved Accuracy**

    Better recognition for major languages

    **Languages**: 18
    **Best for**: High-accuracy needs
  </Card>

  <Card title="🧠 Neural" icon="brain">
    **Highest Quality**

    State-of-the-art neural recognition

    **Languages**: 16
    **Best for**: Premium applications
  </Card>
</CardGroup>

<Callout type="tip">
  **Recommendation**: Start with **Standard** for broad language support, or use **Neural** for English-focused applications requiring the highest accuracy.
</Callout>

***

## Language Support

Azure Speech STT supports 100+ languages and regional variants. Here are the most commonly used:

<Accordion title="Full Language List">
  | Language              | Code    | Standard | Enhanced | Neural |
  | --------------------- | ------- | -------- | -------- | ------ |
  | English (US)          | `en-US` | ✅        | ✅        | ✅      |
  | English (UK)          | `en-GB` | ✅        | ✅        | ✅      |
  | English (Australia)   | `en-AU` | ✅        | ✅        | ✅      |
  | English (Canada)      | `en-CA` | ✅        | ✅        | ✅      |
  | English (India)       | `en-IN` | ✅        | ✅        | ✅      |
  | Spanish (Spain)       | `es-ES` | ✅        | ✅        | ✅      |
  | Spanish (Mexico)      | `es-MX` | ✅        | ✅        | ✅      |
  | French (France)       | `fr-FR` | ✅        | ✅        | ✅      |
  | French (Canada)       | `fr-CA` | ✅        | ✅        | ✅      |
  | German                | `de-DE` | ✅        | ✅        | ✅      |
  | Italian               | `it-IT` | ✅        | ✅        | ✅      |
  | Portuguese (Brazil)   | `pt-BR` | ✅        | ✅        | ✅      |
  | Portuguese (Portugal) | `pt-PT` | ✅        | ✅        | ✅      |
  | Japanese              | `ja-JP` | ✅        | ✅        | ✅      |
  | Korean                | `ko-KR` | ✅        | ✅        | ✅      |
  | Chinese (Mandarin)    | `zh-CN` | ✅        | ✅        | ✅      |
  | Chinese (Hong Kong)   | `zh-HK` | ✅        | ✅        | –      |
  | Chinese (Taiwan)      | `zh-TW` | ✅        | ✅        | –      |
  | Arabic (Saudi Arabia) | `ar-SA` | ✅        | –        | –      |
  | Hindi                 | `hi-IN` | ✅        | –        | –      |
  | Dutch                 | `nl-NL` | ✅        | –        | –      |
  | Russian               | `ru-RU` | ✅        | –        | –      |
  | Swedish               | `sv-SE` | ✅        | –        | –      |
  | Danish                | `da-DK` | ✅        | –        | –      |
  | Norwegian             | `no-NO` | ✅        | –        | –      |
  | Finnish               | `fi-FI` | ✅        | –        | –      |
  | Polish                | `pl-PL` | ✅        | –        | –      |
  | Turkish               | `tr-TR` | ✅        | –        | –      |
  | Hebrew                | `he-IL` | ✅        | –        | –      |
  | Thai                  | `th-TH` | ✅        | –        | –      |
</Accordion>

<Callout type="info">
  **100+ More Languages**: Azure supports many additional languages and regional variants. Visit the [Azure Language Support](https://docs.microsoft.com/azure/cognitive-services/speech-service/language-support) page for the complete list.
</Callout>

***

## Configuration Options

### Basic Configuration

```json theme={null}
{
  "stt_settings": {
    "provider": "azure",
    "model": "standard",
    "language": "en-US"
  }
}
```

### Full Configuration

```json theme={null}
{
  "stt_settings": {
    "provider": "azure",
    "model": "standard",
    "language": "en-US",
    "punctuate": true,
    "interim_results": true,
    "smart_format": true,
    "endpointing": 10,
    "utterance_end_ms": 1000,
    "vad_events": true,
    "keyterms": ["Burki", "AI assistant", "customer support"]
  }
}
```

### Per-Assistant Azure Credentials

You can configure Azure credentials per-assistant instead of using environment variables:

```json theme={null}
{
  "stt_settings": {
    "provider": "azure",
    "azure_config": {
      "subscription_key": "your_subscription_key",
      "region": "eastus"
    }
  }
}
```

***

## Phrase Lists (Keyterms)

Phrase lists boost recognition of specific terms—perfect for company names, product names, and industry terminology.

<Tabs>
  <Tab title="Dashboard">
    In **AI Configuration** → **STT** → **Keywords/Keyterms**:

    Enter terms separated by commas:

    ```
    Burki, AI assistant, voice platform, customer success
    ```
  </Tab>

  <Tab title="API">
    ```json theme={null}
    {
      "stt_settings": {
        "provider": "azure",
        "keyterms": [
          "Burki",
          "AI assistant", 
          "voice platform",
          "customer success"
        ]
      }
    }
    ```
  </Tab>
</Tabs>

<Callout type="tip">
  **Best Practice**: Add your company name, product names, and any domain-specific terminology to phrase lists for improved recognition accuracy.
</Callout>

***

## Key Features

<CardGroup cols={2}>
  <Card title="🎯 Phrase Lists" icon="list">
    **Term Boosting**

    Boost recognition of specific words and phrases for your domain
  </Card>

  <Card title="👥 Speaker Diarization" icon="users">
    **Speaker Identification**

    Distinguish between multiple speakers in a conversation
  </Card>

  <Card title="🔊 Multi-Channel" icon="volume-high">
    **Stereo Support**

    Process audio with separate channels for each participant
  </Card>

  <Card title="⚡ Real-Time" icon="bolt">
    **Low Latency**

    Real-time transcription with interim results
  </Card>
</CardGroup>

***

## Timing Controls

Azure Speech STT supports timing controls to optimize speech detection:

<Accordion title="Endpointing (Silence Threshold)">
  **What it does**: How long to wait after detecting silence before considering speech has ended.

  **Default**: 10ms (minimal endpointing)
  **Range**: 10ms - 2000ms

  **When to Adjust**:

  * **Lower (10-100ms)**: For fast talkers or quick interactions
  * **Higher (500-1000ms)**: For elderly callers or complex topics
  * **Much higher (1500ms+)**: For people with speech difficulties

  ```json theme={null}
  {
    "stt_settings": {
      "endpointing": 500
    }
  }
  ```
</Accordion>

<Accordion title="Utterance End Timeout">
  **What it does**: Maximum time to wait for a complete utterance before triggering end-of-speech.

  **Default**: 1000ms
  **Range**: 500ms - 5000ms

  **When to Adjust**:

  * **Lower (500-800ms)**: For short, quick interactions
  * **Higher (1500-3000ms)**: For detailed conversations

  ```json theme={null}
  {
    "stt_settings": {
      "utterance_end_ms": 1500
    }
  }
  ```
</Accordion>

<Accordion title="VAD Events">
  **What it does**: Enables Voice Activity Detection for enhanced speech detection.

  **Default**: Enabled

  **Benefits**:

  * Better speech detection in noisy environments
  * Backup mechanism when normal detection fails
  * Essential for background noise scenarios

  ```json theme={null}
  {
    "stt_settings": {
      "vad_events": true
    }
  }
  ```
</Accordion>

***

## Provider Comparison

| Feature           | Azure Speech               | Deepgram           |
| ----------------- | -------------------------- | ------------------ |
| **Languages**     | 100+                       | 30+                |
| **Latency**       | \~200ms                    | \~100ms            |
| **Term Boosting** | Phrase Lists               | Keywords/Keyterms  |
| **Diarization**   | ✅                          | ✅                  |
| **Custom Models** | ✅ (Custom Speech)          | Limited            |
| **Real-Time**     | ✅                          | ✅                  |
| **Multi-Channel** | ✅                          | ✅                  |
| **Best For**      | Enterprise, Multi-language | Speed, Phone calls |

<Callout type="info">
  **When to Choose Azure**: Broad language support, Microsoft ecosystem integration, enterprise features, or custom speech models.

  **When to Choose Deepgram**: Ultra-low latency, phone call optimization, or Nova-3 keyterms for English.
</Callout>

***

## Regional Selection

<Callout type="tip">
  **Latency Optimization**: Choose the Azure region closest to your deployment for optimal latency.
</Callout>

| Region          | Location       | Best For                      |
| --------------- | -------------- | ----------------------------- |
| `eastus`        | East US        | North America (East)          |
| `eastus2`       | East US 2      | North America (East) - Backup |
| `westus2`       | West US 2      | North America (West)          |
| `westeurope`    | Netherlands    | Europe                        |
| `northeurope`   | Ireland        | Europe - Backup               |
| `southeastasia` | Singapore      | Asia-Pacific                  |
| `australiaeast` | Australia East | Australia/Oceania             |

***

## Pricing Overview

| Tier         | Hours/Month   | Price              |
| ------------ | ------------- | ------------------ |
| **Free**     | 5 hours       | \$0                |
| **Standard** | Pay-as-you-go | \$1 per audio hour |

<Callout type="info">
  **Enterprise**: Contact Azure for custom pricing on high-volume usage and reserved capacity. Custom Speech model training has separate pricing.
</Callout>

***

## Common Issues & Solutions

<Accordion title="Authentication Failed">
  **Problem**: API returns 401 Unauthorized

  **Solutions**:

  * Verify your Azure Speech Key is correct in **Settings** → **Provider Keys**
  * Ensure the key is from your Speech resource (not another Azure service)
  * Check that the region matches your Speech resource's region
  * Verify your Azure subscription is active
</Accordion>

<Accordion title="Region Mismatch">
  **Problem**: Connection fails or returns errors

  **Solutions**:

  * Double-check the region code (e.g., `eastus` not `east-us`)
  * Ensure the region is available for Speech services
  * Try a different region if experiencing connectivity issues
</Accordion>

<Accordion title="Language Detection Issues">
  **Problem**: Wrong language transcribed

  **Solutions**:

  * Explicitly set the language code in configuration
  * Use the correct regional variant (e.g., `es-ES` vs `es-MX`)
  * Ensure your model supports the selected language
</Accordion>

<Accordion title="Poor Recognition Accuracy">
  **Problem**: Transcription quality is low

  **Solutions**:

  * Add domain-specific terms to phrase lists
  * Try a different model (Enhanced or Neural for supported languages)
  * Ensure audio quality is good (minimal background noise)
  * Enable audio denoising in Burki settings
</Accordion>

<Accordion title="SDK Not Installed">
  **Problem**: Azure Speech SDK import fails

  **Solution**:

  ```bash theme={null}
  pip install azure-cognitiveservices-speech
  ```
</Accordion>

***

## Best Practices

<Cards>
  <Card title="🌍 Regional Selection" icon="globe">
    Deploy in the Azure region closest to your users for minimal latency
  </Card>

  <Card title="📝 Phrase Lists" icon="list">
    Add company names, products, and industry terms for better recognition
  </Card>

  <Card title="🎯 Language Precision" icon="language">
    Use specific regional variants (e.g., `en-GB` vs `en-US`) for best accuracy
  </Card>

  <Card title="⚡ Timing Tuning" icon="clock">
    Adjust endpointing and utterance timeouts based on your caller demographics
  </Card>
</Cards>

***

## See Also

<CardGroup cols={3}>
  <Card title="⚡ Need Speed?" href="/stt-providers">
    **Deepgram** - Ultra-low \~100ms latency, optimized for phone calls
  </Card>

  <Card title="🔧 Timing Controls" href="/stt-providers#advanced-timing-controls">
    **Advanced Settings** - Fine-tune speech detection timing
  </Card>

  <Card title="📞 Call Management" href="/call-management">
    **Conversation Flow** - Configure interruption and timeout settings
  </Card>
</CardGroup>

<Card title="🔗 Additional Resources" icon="link">
  **Azure Speech setup**: [Azure AI Speech speech-to-text quickstart](https://learn.microsoft.com/azure/ai-services/speech-service/get-started-speech-to-text)

  **Language Support**: [Azure STT Language List](https://docs.microsoft.com/azure/cognitive-services/speech-service/language-support)

  **Documentation**: [Azure Speech Service Docs](https://docs.microsoft.com/azure/cognitive-services/speech-service/)

  **Pricing**: [Azure Speech Pricing](https://azure.microsoft.com/pricing/details/cognitive-services/speech-services/)
</Card>

***

<Card title="🚀 Ready to Use Azure Speech STT?" icon="rocket">
  Head back to your [assistant configuration](/ai-configuration) and set up Azure Speech for managed speech-to-text.
</Card>