Speakr — Self-Hosted Audio Transcription Web App

Introduction

Speakr is a self-hosted web application designed for transcribing audio recordings. It runs entirely on your own hardware, keeping sensitive audio data private while providing accurate transcription with speaker identification and an editing interface.

What Speakr Does

• Transcribes audio files using Whisper-based speech recognition models
• Identifies and labels different speakers via diarization
• Provides a web-based editor for reviewing and correcting transcriptions
• Supports multiple audio formats including MP3, WAV, FLAC, and M4A
• Exports transcriptions in various text formats for further use

Architecture Overview

Speakr consists of a Python backend that handles audio processing and transcription using OpenAI Whisper models, combined with a web frontend for file upload and transcript editing. Speaker diarization is handled by dedicated ML models. The application stores transcriptions in a local database and serves everything through a single Docker Compose stack.

Self-Hosting & Configuration

• Deploy with Docker Compose for a one-command setup
• Configure GPU acceleration for faster transcription on NVIDIA hardware
• Select Whisper model size (tiny to large) based on accuracy and speed needs
• Set storage paths for audio files and transcription data
• Configure authentication if exposing the application beyond localhost

Key Features

• Complete privacy with no data leaving your server
• Speaker diarization automatically labels who is speaking
• Web-based editor for correcting and refining transcription output
• GPU acceleration support for fast processing of long recordings
• Multi-format audio input with flexible text export options

Comparison with Similar Tools

• Whisper (CLI) — raw command-line transcription; Speakr adds a web UI, diarization, and editing
• Otter.ai — cloud-based transcription service; Speakr is self-hosted and private
• MacWhisper — macOS-only desktop app; Speakr runs on any platform via Docker
• Buzz — desktop Whisper GUI; Speakr is web-based and supports multiple users
• WhisperX — CLI tool with alignment; Speakr wraps similar capabilities in a web interface

FAQ

Q: Do I need a GPU? A: No, but a GPU significantly speeds up transcription. CPU-only mode works with smaller Whisper models.

Q: How accurate is the transcription? A: Accuracy depends on the Whisper model size chosen. The large model provides professional-grade results for clear audio.

Q: Can multiple users share the same instance? A: Yes. The web interface supports concurrent access for teams sharing a self-hosted instance.

Q: What languages are supported? A: Speakr supports all languages that Whisper supports, which covers over 90 languages.

Sources

• https://github.com/murtaza-nasir/speakr