Keywords Boost

Improve recognition accuracy for specific words, names, or domain-specific terminology.

Overview

Keywords boost helps the model better recognize specific terms that might otherwise be misheard. This is especially useful for:

• Product names and brand names
• Company names and person names
• Technical jargon and acronyms
• Industry-specific vocabulary

How It Works

Speech-to-text models sometimes misrecognize uncommon words:

• Brand names: "YourVoic" → "Your Voice"
• Technical terms: "webhook" → "web hook"
• Acronyms: "API" → "a pie"

By providing a list of keywords, you tell the model: "These exact words might appear - listen for them!"

The model then boosts the probability of recognizing those specific terms when it hears something similar:

1. You pass keywords: "YourVoic,API,webhook"
2. The STT model adds these to its vocabulary with higher weight
3. When audio sounds like "your voice", the model checks: "Could this be 'YourVoic'?"
4. If close enough, it transcribes as "YourVoic" instead of "Your Voice"

Supported Models

Usage

Pass keywords as a comma-separated string:

import requests

response = requests.post(
    "https://yourvoic.com/api/v1/stt/lucid/transcribe",
    headers={"X-API-Key": "your_api_key"},
    files={"file": open("interview.mp3", "rb")},
    data={
        "model": "lucid-mono",
        "keywords": "YourVoic,API,transcription,Lucid,Cipher"
    }
)

result = response.json()
print(result["text"])

Real-time Streaming

Keywords can also be used in real-time WebSocket connections:

const params = new URLSearchParams({
    api_key: 'YOUR_API_KEY',
    model: 'lucid-agent',
    keywords: 'YourVoic,API,webhook,streaming'
});

const ws = new WebSocket(`wss://yourvoic.com:8443/api/v1/stt/realtime/stream?${params}`);

Best Practices

• Limit keywords: Use 10-20 keywords maximum for best performance
• Exact spelling: Provide keywords exactly as you want them transcribed
• Include variations: Add common misspellings if relevant
• Case sensitivity: Keywords are case-insensitive

Context Prompts (Cipher Alternative)

For Cipher models, use the prompt parameter instead:

response = requests.post(
    "https://yourvoic.com/api/v1/stt/cipher/transcribe",
    headers={"X-API-Key": "your_api_key"},
    files={"file": open("audio.mp3", "rb")},
    data={
        "model": "cipher-max",
        "prompt": "Domain vocabulary: YourVoic, API, transcription, webhook"
    }
)