|
| 1 | +# Deepgram Keyword Boosting |
| 2 | + |
| 3 | +Deepgram’s keyword boosting feature allows further customization to improve the transcription accuracy. This is particularly useful in customer service scenarios where specialized terminology, product names, or customer names are frequently misrecognized. |
| 4 | + |
| 5 | +## Why Use Keyword Boosting? |
| 6 | +Keyword boosting is beneficial for: |
| 7 | + |
| 8 | +- Enhancing the recognition of customer names and brand-specific terms. |
| 9 | +- Improving transcription accuracy in outbound sales, inbound support, and ordering systems. |
| 10 | +- Reducing errors in transcriptions of product or service names. |
| 11 | + |
| 12 | +## API Call Integration |
| 13 | +To enable keyword boosting, add a `keywords` parameter to your Vapi assistant’s `transcriber` section. This should include the keywords and their respective intensifiers. |
| 14 | + |
| 15 | +### Example of POST Request |
| 16 | +```bash |
| 17 | +curl \ |
| 18 | + --request POST \ |
| 19 | + --header 'Authorization: Bearer <token>' \ |
| 20 | + --header 'Content-Type: application/json' \ |
| 21 | + --data '{ |
| 22 | + "name": "PizzaBot", |
| 23 | + "model": { |
| 24 | + "model": "gpt-4o", |
| 25 | + "provider": "openai" |
| 26 | + }, |
| 27 | + "voice": { |
| 28 | + "voiceId": "pizza-assistant", |
| 29 | + "provider": "azure" |
| 30 | + }, |
| 31 | + "transcriber": { |
| 32 | + "provider": "deepgram", |
| 33 | + "model": "nova-2", |
| 34 | + "language": "en", |
| 35 | + "smartFormat": true, |
| 36 | + "keywords": [ |
| 37 | + "Margherita:3", |
| 38 | + "Pepperoni Deluxe:3", |
| 39 | + "BBQ Chicken:3" |
| 40 | + ] |
| 41 | + }, |
| 42 | + "firstMessage": "Welcome to Pizza Express! What would you like to order?", |
| 43 | + "firstMessageMode": "assistant-speaks-first" |
| 44 | + }' \ |
| 45 | + https://api.vapi.ai/assistant |
| 46 | +``` |
| 47 | + |
| 48 | +### Explanation of Parameters: |
| 49 | +- `name`: The name of the assistant. |
| 50 | +- `model`: Specifies the model and provider for the assistant’s conversational capabilities. |
| 51 | +- `voice`: Specifies the voice and provider for the assistant’s speech. |
| 52 | +- `transcriber`: Uses Deepgram for transcription, with custom keywords for pizza menu items. |
| 53 | +- `firstMessage`: The initial message the assistant will speak. |
| 54 | +- `firstMessageMode`: Specifies that the assistant speaks first. |
| 55 | + |
| 56 | +## Intensifiers |
| 57 | +Intensifiers adjust how strongly the model prioritizes certain words: |
| 58 | + |
| 59 | +- **Boosting Example:** `keywords=Margherita:4` (higher likelihood of recognition) |
| 60 | +- **Suppressing Example:** `keywords=Anchovies:-2` (lower likelihood of recognition) |
| 61 | + |
| 62 | +## Common Use Cases |
| 63 | +### 1. Pizza Order System |
| 64 | +Imagine you’re building a voice agent for a pizza delivery service. Customers frequently order pizzas with specific names that may not be easily recognized by default speech models. You can use keyword boosting to ensure accurate transcription of menu items like "Margherita," "Pepperoni Deluxe," or "BBQ Chicken." |
| 65 | + |
| 66 | +### 2. Financial Services |
| 67 | +In banking and finance, customers may mention account types, transaction codes, or financial products. Using keyword boosting ensures that terms like "ACH Transfer," "FICO Score," or "401k" are transcribed accurately. |
| 68 | + |
| 69 | +### 3. Healthcare Support |
| 70 | +Medical professionals and patients might use technical medical terms during calls. Boosting keywords such as "Amoxicillin," "MRI Scan," or "Hypertension" helps maintain accuracy in transcription. |
| 71 | + |
| 72 | +### 4. Travel and Hospitality |
| 73 | +When booking flights or hotels, recognizing destination names and travel jargon correctly is crucial. Boosting city names like "Honolulu," "Santorini," or "Maldives" ensures better accuracy. |
| 74 | + |
| 75 | +## Best Practices for Keyword Boosting |
| 76 | +- **Use Customer-Focused Keywords:** Common product names, store locations, and customer names. |
| 77 | +- **Set Balanced Intensifiers:** Avoid extreme values that could cause false positives. |
| 78 | +- **Limit Excessive Keywords:** Overloading with too many keywords may reduce effectiveness. |
| 79 | + |
| 80 | +By implementing keyword boosting, customer service applications can significantly improve transcription accuracy, leading to better user experiences. |
0 commit comments