murtaza-nasir/speakr

What it solves

Speakr is a self-hosted platform designed to transform audio recordings into organized, searchable, and intelligent notes while maintaining complete privacy by running on the user's own infrastructure.

How it works

The platform uses a pipeline consisting of four main stages:

1. Capture: Audio is recorded via microphone, system/browser audio, or imported via file upload and a watched folder.
2. Transcribe: It integrates with various ASR (Automatic Speech Recognition) engines, including self-hosted WhisperX (for speaker diarization and voice profiles) or cloud APIs like OpenAI, Mistral, and AssemblyAI.
3. Understand: It uses LLMs to generate automatic summaries, extract action items and calendar events, and provide a chat interface for querying individual recordings or the entire library (Inquire Mode).
4. Organize: Users can manage recordings using folders, smart tags (which can carry their own AI prompts), and retention policies for auto-deletion.

Who it’s for

It is built for privacy-conscious individuals and groups, including families preserving memories, students taking lecture notes, professional teams managing project discussions, and legal or sales teams requiring specific retention and sharing policies.

Highlights

• Flexible Transcription: Support for multiple backends including self-hosted WhisperX for high-quality diarization and voice embeddings.
• Intelligent Note-taking: Custom prompt-based tags that can be stacked to transform raw transcripts into specific formats like recipes or study notes.
• Privacy-First: Entirely self-hosted with support for Single Sign-On (OIDC) and granular sharing controls.
• Semantic Search: "Inquire Mode" allows natural-language chatting and searching across the entire audio library.
• Automation: Includes a REST API and signed webhooks for integration with tools like n8n, Zapier, and Make.