Applaud

Applaud is free, private, and open source audio transcription and summarization tool. It's designed to take audio recordings and provide a transcription, as well as common AI use-cases like flashcards, questions, and summaries.

The goal is to provide a self-hosted alternative to https://www.plaud.ai/. Some people don't like the idea of uploading their audio recordings to a third-party service. With Applaud, you can host it yourself and keep your data private. Applaud even supports local LLM models using Ollama.

NOTE: This is not intended to be hosted on a public server. It is designed to be run locally on your own machine. There is no authentication or authorization built into the application.

Features

• Automatically syncs audio recordings (e.g. iCloud Drive, Google Drive, etc.)
• Transcribes audio recordings into a JSON file using insanely-fast-whisper with full CUDA and MPS (Apple Silicon) support
• Summarizes the transcript using the models of your choice
• Generates flashcards, questions, and answers from the transcript

Supported LLM Providers

✅ OpenAI (.env OPENAI_API_KEY)

✅ Anthropic (.env ANTHROPIC_API_KEY)

✅ Google (.env GOOGLE_API_KEY)

✅ Ollama (.env OLLAMA_BASE_URL and OLLAMA_API_KEY)

✅ OpenRouter (.env OPENROUTER_API_KEY)

Once the frontend and backend are deployed, visiting http://localhost:3000 for the first time will prompt you to set up your LLM provider. You can change models at any time by selecting the cog icon in the top right corner. If you wish to use Ollama, there are a few additional steps to take.

Ollama Setup

1. Install ollama and start the server: ollama serve
2. Create a 120k context window model
   • ollama install llama3.2
   • ollama show --modelfile llama3.2 > Modelfile
   • Set PARAMETER num_ctx 120000
   • ollama create -f Modelfile llama3.2-120k
3. Configure the .env OLLAMA_BASE_URL to point to your Ollama instance (typically http://localhost:11434/v1 unless you're hosting it on an external server)
4. Configure the .env OLLAMA_API_KEY to your Ollama API key (usually doesn't matter, but if you have secured your Ollama instance, you may need to set this)
5. Visit the frontend and choose ollama as the provider and llama3.2-120k as the model

Requirements

This has been tested on macOS. Linux support should be supported. Windows support is not guaranteed.

• docker compose
• python@3.11
• ffmpeg

Setup

1. Clone the repository: git clone https://github.com/landoncrabtree/applaud.git && cd applaud
2. Prepare the environment variables: cp .env.example .env
3. Refer to the watcher/README.md for instructions on how to setup and start the watcher service
4. Modify .env with any API keys for different LLM providers
5. Start the frontend and backend services: docker compose up
   • The frontend will be available at http://localhost:3000
   • The backend will be available at http://localhost:8080

Screenshots

Benchmarks

Using insanely-fast-whisper with distil-whisper/large-v2 and pyannote/speaker-diarization-3.1 on a M2 Max MacBook Pro (32GB RAM, 30 GPU Cores)

Audio Duration	File Size	Transcription Time	Segmentation Time (speaker diarization)
54:56	67.6MB	09:48	02:36

TODO

• Fix watcher sometimes duplicating files
• UI/UX improvements
• Add more AI tools
• Chat side panel with conversation history per transcript
• Change all hardcoded references to localhost:8080 to be dynamic
• Fix delete foreign key constraint (just delete all references first, then delete the transcript)