Scriber

Turn live speech, YouTube videos, and media files into useful text from one Windows workspace.

Dictate into any app. Keep every transcript searchable. Choose the cloud or local model that fits the job.

One desktop. Every spoken workflow.

Scriber is not another upload page wrapped around one speech API. It is a Windows-first transcription workspace for the moments before, during, and after speech becomes text.

	Typical transcription tools	Scriber
Live dictation	Text stays inside the recorder	A global hotkey records, transcribes, and pastes into the app you are already using
Input sources	Meetings, files, or YouTube as separate products	Meetings, live microphone, YouTube, and local media share one workspace
Model choice	One vendor decides quality, latency, and price	Switch between streaming, async, batch, and local ONNX providers
YouTube	Always download and transcribe audio	Prefer captions first, then fall back to audio automatically
After transcription	Copy raw text and leave	Search, summarize, post-process, export, copy, and reopen later
Desktop integration	Browser tab and manual uploads	Native tray, hotkeys, autostart, overlay, updater, and supervised backend
Troubleshooting	Generic error message	Filtered runtime logs, redacted diagnostics, and support bundles

Built around the way you capture information

🎙️ Dictate into any Windows app

Use the normal hotkey for fast, faithful speech-to-text. Use the separate post-processing hotkey when the result should be cleaned, structured, and ready to send.

• Ready-to-use defaults on a fresh install: Ctrl+Shift+D for Live Mic and Ctrl+Shift+F for post-processing; existing Settings or .env choices are preserved
• Rust and WASAPI microphone capture
• Optional microphone pre-warming for lower startup latency
• Native recording overlay with live feedback
• Raw dictation or prompt-driven post-processing
• Clipboard-aware insertion with bounded clipboard restoration
• Searchable recent recordings with useful transcript excerpts

▶️ YouTube captions when possible. Audio when needed.

Paste a video URL or search YouTube inside Scriber. Creator captions and automatic captions are tried first, which can make a transcript available faster and avoid unnecessary provider cost. If captions are unavailable, Scriber prepares the audio and continues automatically.

• Caption-first behavior is configurable under Settings > Summarization
• Current yt-dlp extraction with EJS and a bundled offline QuickJS-ng wrapper
• Bundled ffmpeg and ffprobe media preparation
• Durable progress, retry, cancel, and recovery state
• Transcript and summary saved beside every other source

📁 Bring recordings in without upload gymnastics

Drop in audio, video, or several files at once. Scriber extracts audio, compresses large inputs when useful, tracks the queue, and keeps completed work organized.

• MP3, M4A, WAV, MP4, MOV, and other common media formats
• Multi-file batch import
• Automatic audio extraction from video
• Provider-aware preparation for large inputs
• Progress, cancellation, retry, and durable job state
• PDF and DOCX export from completed transcripts

👥 Capture meetings without a joining bot

The Meetings workspace records microphone and Windows system audio locally while the call is in progress. WebRTC AEC3 uses the system-audio reference to remove speaker echo from the microphone track; the raw source is still retained for recovery. After stop, Scriber creates a timestamped canonical transcript, summary, decisions, action items, cited chat answers, and reusable exports.

• Independent raw mic, AEC-clean mic, and system tracks on one timeline
• Crash-safe chunks, pause/resume gaps, startup recovery, and configurable audio retention
• Provider-native timestamps and speaker turns where supported
• Optional offline Sherpa-ONNX speaker separation for STT models without native diarization; the signed app supplies an isolated static worker while only the models are downloaded, shared by File, YouTube, Meetings, and Meeting imports
• Import an existing audio or video recording directly into the Meeting workspace
• Ctrl+Shift+M opens, restores, and focuses the Meeting workspace on a fresh install; an existing custom shortcut remains unchanged
• Optional local speaker library with explicit biometric opt-in and deletion
• Optional Outlook Calendar connection with a refreshable list of today's events, organizer and attendee details, and an explicit event/no-event choice; the selected details are frozen with the Meeting and no bot joins the call. Scriber shows when the calendar was last refreshed, warns if a selected event was moved or removed, and offers a clear reconnect path when Microsoft access expires.
• Voice Library and local-account matches are suggested first after the Meeting; optional AI suggestions for unresolved speakers always require confirmation
• Name unresolved speakers directly in the Meeting, including people, teams, rooms, or shared microphones. These meeting-local labels do not create Outlook identities, rename Voice Library profiles, or add email recipients
• Permanently merge duplicate speaker identities when one person was detected as two speakers; Scriber preserves explicitly assigned Meeting labels and asks for confirmation before changing the local Voice Library
• Email drafts reuse the selected event's suitable participant addresses and show recipients for review before sending
• Markdown, JSON, PDF, DOCX, multitrack FLAC, and synchronized Opus playback

✨ The transcript is the beginning, not the result

Every source lands in the same local transcript library. From there, Scriber helps turn raw speech into something you can actually use.

• Search across live recordings, YouTube videos, and imported files
• Automatic or manual summaries
• Follow structured File and YouTube summaries with a scroll-synchronized table of contents whose label follows the summary language
• Separate models and prompts for summaries and live post-processing
• Copy transcript or summary independently
• Export polished results to PDF or DOCX
• Preserve speaker labels for supported diarized batch providers
• Reopen processing, failed, stopped, summary-failed, and completed jobs with clear state

Choose the model. Keep the workflow.

Latency, accuracy, price, privacy, and language support vary by task. Scriber keeps the workflow stable while you choose the provider.

Cloud streaming

Use a realtime provider when words should appear while you speak.

Cloud async and batch

Use completed-audio processing for long recordings, file imports, and providers optimized for accuracy or cost.

Modulate.AI multilingual transcription

Modulate.AI is available for multilingual batch and realtime transcription. Scriber keeps only final transcript text: partial results and optional emotion, accent, deepfake, and PII/PHI signals are disabled. The provider's published base prices are $0.03 per audio hour for batch and $0.06 per audio hour for streaming, without optional add-ons. Settings uses 4.43% word error rate as the comparison benchmark and sorts models by their displayed error rate. See Modulate's official API pricing. Transcription quick start.

Soniox US and EU regions

Soniox defaults to its US region. Open the Soniox API-key popup in Settings to switch both realtime and uploaded-audio transcription to the EU region. EU access must first be enabled for your Soniox organization: contact support@soniox.com, create an EU project, and use that project's region-specific API key. The region selection and key must match; a US project key does not provide EU data residency. Read the official Soniox data-residency guide.

Local ONNX

Download supported ONNX models from Settings and transcribe locally without an STT API key. Scriber uses prepared model artifacts rather than asking end users to install or run NeMo and Torch.

Current provider coverage includes Soniox, Modulate.AI, AssemblyAI, Microsoft Azure MAI, OpenAI, OpenRouter, Celeris, Deepgram, Mistral, Gladia, Groq, Speechmatics, Smallest AI, ElevenLabs, Gemini, Google Cloud, and ONNX.

🔑 Credentials and AI behavior stay understandable

Provider credentials, transcription models, summary models, prompts, language behavior, and update controls live in one Settings workspace. Unavailable cloud models remain visibly gated until the matching credential exists.

• Credential status and direct provider links
• Separate STT, summarization, and post-processing model choices
• Practical price and error estimates where benchmark data is available
• Custom vocabulary for names, brands, and domain language
• Automatic summarization and caption-first controls
• Gemini, OpenRouter, OpenAI, Cerebras, and Celeris summary paths
• Light, dark, and system theme support
• Complete German and English interface with a persistent language switch in the app shell and Settings; transcription-language choices remain separate from the interface language

🛡️ Local-first where it matters

Scriber runs its desktop shell, frontend, backend, transcript database, settings, and history on your machine.

Data	What happens
Transcript history	Stored locally in Scriber's user data directory
Microphone and media	Sent only when the selected transcription provider is cloud-based
Summaries and post-processing	Text is sent only to the model provider you choose for that action
Local ONNX transcription	Runs without an STT API key
Credentials	Stored for the configured provider and redacted from support diagnostics
Support bundles	Known API keys, bearer tokens, and session secrets are redacted

Scriber does not require a Scriber account. Cloud providers still apply their own pricing, retention, and privacy terms.

🧰 A debugging console you can actually use

Transcription crosses microphones, media tools, model APIs, networking, and desktop integration. Scriber makes those boundaries visible without exposing raw transcript content in diagnostics.

• Severity, source, date, component, and message filters
• Clear selected view or persisted runtime logs
• Copy visible diagnostics
• Generate a redacted support bundle
• Inspect post-processing health without logging transcript text
• Track live runtime state and provider failures

Windows-native by design

The installed app is more than a packaged website:

• Tauri 2 shell for tray actions, autostart, global shortcuts, single-instance behavior, updates, and backend supervision
• Rust audio sidecar for crash-isolated WASAPI microphone capture and pre-warming
• React workspace for fast navigation across Live Mic, YouTube, File, Settings, Console, and transcript details
• Python backend sidecar for provider routing, job state, summaries, local storage, and support tooling
• Bundled media stack with ffmpeg, ffprobe, yt-dlp extraction support, and a bounded QuickJS-ng runtime
• Signed updater artifacts with published checksums and release diagnostics

Download and start

1. Download the latest Windows installer.
2. Open Settings and choose a transcription provider.
3. Add the matching API key, or download a local ONNX model.
4. Start with a short Live Mic recording, then try YouTube or a media file.

The standard installer includes the desktop shell, frontend, backend sidecar, Rust audio sidecar, and media tools needed for normal use.

Windows is the primary supported platform. Linux and macOS paths are intended mainly for development and fallback use today.

Frequently asked questions

Can I use Scriber without an API key?

Yes. Choose a supported local ONNX transcription model. Cloud transcription, summarization, and post-processing require credentials for the provider you select.

Does YouTube transcription always download the video?

No. Caption-first mode checks useful creator and automatic caption tracks first. Audio is prepared only when captions are disabled or unavailable.

Can I use different models for transcription and summaries?

Yes. Speech-to-text, summarization, and live post-processing have separate model and prompt choices.

Where are my transcripts stored?

Transcript history and runtime data stay in Scriber's local user data directory. Selected cloud features send only the audio or text required by that provider action.

What should I try first?

For a simple cloud setup, configure a Gemini API key for Gemini STT and summaries. For local transcription, download an ONNX model from Settings.

For developers

Architecture, setup, build, and test commands

Requirements

• Windows 10 or newer for the primary desktop runtime
• Python 3.13
• Node.js 26.3.1 and npm
• Stable Rust toolchain
• Git

Start the desktop app in development

git clone https://github.com/MyButtermilk/Scriber.git
cd Scriber\Frontend
npm install
npm run tauri:dev

Run the backend and frontend separately when you are working on one layer:

cd Scriber
py -3.13 -m venv venv
venv\Scripts\python.exe -m pip install -r requirements.txt
scripts\project-python.cmd -m src.web_api

cd Scriber\Frontend
npm install
npm run dev:client

Build a Windows installer

powershell -NoProfile -ExecutionPolicy Bypass -File scripts\build_windows.ps1 `
  -FastLocalInstaller `
  -RunInstallerFrontendSmoke `
  -RunInstallerMediaPreparationSmoke

The NSIS installer is written to:

Frontend\src-tauri\target\release\bundle\nsis\

Test

scripts\project-python.cmd -m pytest

cd Frontend
npm run check
npm run build

cd Frontend\src-tauri
cargo test

Run the real-browser frontend smoke against its privacy-safe synthetic backend:

scripts\project-python.cmd scripts\smoke_frontend_browser.py --output tmp\frontend-browser-smoke.json

Active documentation

• AGENTS.md: repository editing guide and non-negotiable contracts
• docs/ARCHITECTURE.md: runtime architecture and ownership boundaries
• docs/PERFORMANCE_AND_PACKAGING.md: performance and installer decisions
• docs/TESTING_AND_RELEASE.md: smoke gates, signing, updater, and release flow
• docs/ROADMAP_AND_KNOWN_ISSUES.md: prioritized next work and open issues

---

Speech becomes valuable when it can move.
Capture it once. Search it, shape it, and use it anywhere.