Categories
Self-Hosted Subtitle Generation AI Summaries Audio Transcription Speech-to-Text Speaker Diarization Meeting TranscriptsStars
Forks
Watchers
Developer links
Scriberr
When a meeting, interview, lecture, or YouTube video needs a transcript, Scriberr produces a searchable, speaker-labeled one on a server you run. Upload audio or video. Paste a YouTube link. Record in the browser. Import a multi-track Audacity project. Auto-transcribe starts every new upload with your default profile. Parakeet, Canary, and Whisper models run locally with word-level timestamps, dozens of languages, and optional translation into English. Speaker labels come from pyannote or NVIDIA Sortformer diarization, after the models are fetched once with a Hugging Face token. In the reader, each word lights up as the audio plays, and Cmd or Ctrl click jumps to any word. A switch flips between a timeline and compact text. Highlight a passage to attach a time-stamped note. Export JSON, SRT subtitles, or plain text. Point Scriberr at Ollama or any OpenAI-compatible endpoint to summarize with your own prompt templates or chat with a transcript. Saved profiles remember language, model, and prompt settings. Bulk mode queues or deletes many recordings at once. API keys, webhook callbacks, and a folder-watching command line client let n8n flows and scripts submit audio and collect transcripts. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.
Benefits
- Recordings Stay on Your Server
- Transcription and diarization run on local models on your own VPS, so interviews, meetings, and voice notes leave it only if you choose OpenAI for transcription or chat.
- Unlimited Hours Without Subscription Fees
- Processing happens on your own hardware, so hours of audio cost only the VPS; the author built Scriberr after a cloud plan priced at $100 a year for 20 monthly hours.
- Repeatable Settings for Every Upload
- Transcription profiles store model, language, and prompt choices, and auto-transcribe applies the default profile to every new upload, so recurring recordings need no manual setup.
- Runs on CPU, Faster on GPU
- A CPU Docker image runs on servers without a graphics card, and separate CUDA images use an NVIDIA GPU to speed up processing.
Features
- Local Speech Models
- Runs NVIDIA Parakeet, NVIDIA Canary, and Whisper models on your hardware with word-level timestamps, dozens of languages, and optional translation into English during transcription.
- Speaker Diarization
- Labels who said what using pyannote or NVIDIA Sortformer, running locally once a Hugging Face token has downloaded the models and the user agreements are accepted.
- Synchronized Playback and Notes
- Highlights each word as audio plays, seeks on Cmd or Ctrl click, switches between timeline and compact views, and attaches time-stamped notes to selected text.
- Summaries and Transcript Chat
- Connects to Ollama or any OpenAI-compatible endpoint, generates summaries from reusable prompt templates, and opens chat sessions for asking questions about a recording.
- Flexible Inputs and Exports
- Accepts audio and video files, YouTube links, browser recordings, and multi-track Audacity projects, then downloads transcripts as JSON, SRT subtitles, or plain text.
- REST API and Webhooks
- API keys, a callback URL on each job, and a folder-watching command line client let n8n flows and scripts submit audio and receive finished transcripts automatically.