wavecap-llm

What problem does it solve?

This Skill addresses inaccuracies and jargon in automated transcriptions by leveraging local Large Language Models (LLMs) for intelligent correction, ensuring higher fidelity and domain-specific accuracy.

Core Features & Use Cases

• LLM-based Correction: Automatically corrects errors in Whisper transcriptions using configurable local LLMs.
• Model & Prompt Tuning: Allows users to select different LLM models, adjust generation parameters (temperature, max tokens), and define domain-specific terms to preserve jargon.
• Use Case: A medical professional needs highly accurate transcriptions of patient consultations. This Skill can be configured with a suitable LLM and domain terms like medical abbreviations to ensure the final transcriptions are precise and professional.

Quick Start

Configure the wavecap-llm skill to enable LLM correction using the 'llama-3.2-3b' model and set the temperature to 0.1.