Libraries

Software

About

Contact

Menu

Close

Libraries

Software

About

Contact

Menu

Close

SM-Ai: Audio Metadata AI Processor

Overview

SM-Ai is a macOS application designed to enhance and standardize audio file metadata using artificial intelligence. The application processes audio metadata from Soundminer databases, leveraging large language models to analyze, categorize, and reformat metadata.

Key Features

  • AI-Powered Metadata Processing: Uses OpenAI GPT models, Anthropic Claude models, or local LLMs to intelligently process and enhance audio metadata

  • UCS Category Intelligence: Built-in Universal Category System support with 850+ predefined categories for accurate sound classification

  • Flexible Output: Write AI-generated content to any metadata field with options to append, prepend, or replace existing content

  • Batch Processing: Handle multiple files efficiently through Soundminer's batch processing capabilities

  • Custom Prompts: Create and save custom prompt templates for different metadata enhancement workflows


How It Works

  1. Select audio files in Soundminer

  2. Run the included Lua script to send metadata to SM-Ai

  3. SM-Ai processes the metadata through your chosen AI model

  4. Enhanced metadata is written back to your specified field in Soundminer


System Requirements

  • macOS 13.5 or later

  • Soundminer V4.5Pro, V5Pro,  V6 Pro or V6 Plus

  • Internet connection (for cloud AI providers) or local LLM setup


API Setup

SM-Ai supports three types of AI providers: OpenAI, Anthropic Claude, and local LLMs via Ollama.

OpenAI Setup

OpenAI provides the GPT family of models (GPT-4o, GPT-4o-mini, GPT-3.5-turbo).

  1. Create an Account
    Visit: https://platform.openai.com/signup

  2. Generate an API Key

    • Log in to your account

    • Navigate to: https://platform.openai.com/api-keys

    • Click "Create new secret key"

    • Give it a descriptive name (e.g., "SM-Ai Metadata Processing")

    • Copy the key immediately (you won't be able to see it again)

  3. Add Billing Information

  4. Enter Key in SM-Ai

    • Open SM-Ai Settings (⌘,)

    • Select an OpenAI model from the AI Provider dropdown

    • Paste your API key in the API Key field

Anthropic Claude Setup

Anthropic provides the Claude family of models (Claude Opus, Sonnet, and Haiku).

  1. Create an Account
    Visit: https://console.anthropic.com/signup

  2. Generate an API Key

  3. Add Credits

  4. Enter Key in SM-Ai

    • Open SM-Ai Settings (⌘,)

    • Select a Claude model from the AI Provider dropdown

    • Paste your API key in the API Key field

Local LLM Setup (Ollama)

For offline processing and/or cost savings, you can use loc al language models through Ollama.

  1. Download and Install Ollama
    Visit: https://ollama.com/download

  2. Install a Model
    You can install a new model using either Terminal or the GUI chat window.

    • Terminal mode: ollama pull llama3.1

    • GUI mode: Choose a model in Ollama’s chat app and it will download and install the chosen model once you initiate a chat (ie. type “hello”)

  3. Start Ollama Server
    Ollama runs automatically in the background after installation. You can verify it's running by type into Terminal:
    ollama list

  4. Configure SM-Ai

    • Open SM-Ai Settings (⌘,)

    • Select "Local LLM" from the AI Provider dropdown

    • Chose from downloaded models 

    • Verify the Base URL is: http://localhost:11434/v1/chat/completions

Note: Local LLMs require significant system resources. For best performance:

  • Apple Silicon Macs (M1/M2/M3) with 16GB+ RAM recommended

  • Larger models (70B+) require 32GB+ RAM

  • Processing will be slower than cloud APIs but is free and private


Cost Considerations

Cloud AI Pricing (Approximate as of October, 2025)

OpenAI:

  • GPT-4o: ~$2.50 per 1,000 prompts (typical metadata processing)

  • GPT-4o-mini: ~$0.15 per 1,000 prompts

  • GPT-3.5-turbo: ~$0.50 per 1,000 prompts

Anthropic:

  • Claude Opus 4.1: ~$15 per 1,000 prompts

  • Claude Sonnet 4/4.5: ~$3 per 1,000 prompts

  • Claude Haiku 3.5: ~$0.80 per 1,000 prompts

Local LLM:

  • Free (no per-use cost)

Actual costs vary based on prompt length and model responses. Monitor your usage through the respective provider dashboards.


Getting Started

  1. Install SM-Ai: Launch the application and complete the initial setup

  2. Configure API: Follow the API setup instructions above for your chosen provider

  3. Install Soundminer Script: Copy the included Send_to_SM-Ai.lua script to your Soundminer Scripts folder (typically ~/Library/Application Support/Soundminer/Scripts/)

  4. Start Processing: Select files in Soundminer and run the script


Learning Prompt Engineering

Effective prompt engineering is key to getting the best results from AI models. These resources will help you understand how to craft better prompts:

General Prompt Engineering

Advanced Techniques

Key Concepts to Learn

  • Clear Instructions: Be explicit about what you want

  • Context Provision: Give the AI relevant background information

  • Examples: Show the AI what good output looks like

  • Output Format: Specify exactly how you want results structured

  • Iterative Refinement: Start simple and progressively improve your prompts

  • System Messages: Use system prompts to set consistent behavior


Tips for Best Results

  1. Be Specific: The more detailed your prompts, the better the results

  2. Provide Context: Include relevant information about your audio library's organization

  3. Use Examples: Show the AI examples of good descriptions from your library

  4. Test First: Process a small batch before running on your entire library

  5. Monitor Costs: Check your API usage regularly, especially when starting out

  6. Backup: Always backup your database before batch processing


Privacy & Data

  • Cloud AI: Your metadata is sent to OpenAI or Anthropic servers for processing. Review their privacy policies before use.

  • Local LLM: All processing happens on your computer; no data is sent externally.

  • SM-Ai: Does not collect or transmit any user data