SpeechForge

SpeechForge turns audio into editable transcripts, translates text with DeepL, and lets you continue working with a transcript in Mistral chat. The same React application targets the web, Linux/Windows desktop through Tauri v2, and Android through Capacitor.

Highlights

• Upload or record audio and transcribe it with Mistral AI.
• Preserve the native Linux/Tauri recording path while using MediaRecorder on web and Android.
• Translate text with DeepL language, formality, model, sentence-splitting, and formatting controls.
• Attach the current transcript as private in-session chat context without exposing the system prompt in the UI.
• Switch between Transcribe, Translate, and Chat without losing current work; destructive Home resets require confirmation.
• Use the guided provider setup or configure Mistral and DeepL later in Settings.
• Choose English or Italian and system, light, or dark appearance.
• Run an accessible Material 3 interface across desktop and mobile layouts.

Getting Started

Requirements:

• Bun 1.3 or newer
• Rust and the Tauri v2 platform dependencies for desktop builds
• Android SDK, command-line tools, and Java 21 for Android builds
• Docker for the containerized web target

Install and start the web app:

bun install
bun run dev

The first-run setup explains each workspace and can validate one or both providers. Mistral is required for transcription and chat; DeepL is required for translation. Settings validates every configured provider before committing any edited API setting, so a partial validation failure does not overwrite the last working configuration. Credentials remain in local device storage and are sent only to the configured provider when used.

Commands

Action	Command	Description
Install	bun install	Install dependencies from bun.lock.
Web development	bun run dev	Start the hot-reloading web server.
Web build	bun run build	Build production assets into dist/.
Tests	bun test	Run service, hook, context, and component tests.
Lint	bun run lint	Run ESLint over TypeScript and React sources.
Typecheck	bunx --bun tsc --noEmit	Run strict TypeScript validation.
Tauri development	bun run dev:tauri	Start the desktop application in development mode.
Tauri build	bun run build:tauri	Build the native desktop application.
Android sync	bun run cap:sync	Build web assets and sync them into Android.
Temporary ADB run	bun run build:adb	Build/install the debug app, then uninstall it when stopped.

bun run build:adb uses com.speechforge.app.debug, so it can run beside the release app. Signed release artifacts are produced by the tag-based GitHub Actions workflows.

Project Structure

• src/components/: app workspaces, shell components, provider setup, and reusable UI primitives
• src/context/: locale and theme state
• src/hooks/: transcription, dual-path recording, provider settings, and shared-file workflows
• src/i18n/: typed English and Italian messages
• src/services/audio/: duration, normalization, and long-audio splitting
• src/services/mistral/ and src/services/deepl/: provider clients and platform-specific transport
• src-tauri/: Tauri configuration and Rust commands for desktop recording/network access
• android/: Capacitor Android project
• test/: Happy DOM and Testing Library setup for Bun component tests
• build.ts and dev.ts: custom Bun/Tailwind build pipeline

Platform Notes

• Linux inside Tauri uses the Rust native recorder. Browser, Windows webview, and Android recording use the web MediaRecorder path.
• DeepL calls use the web proxy in browser builds, a Rust command in Tauri, and CapacitorHttp on Android.
• Run bun run cap:sync before every Android build so android/app/src/main/assets/public/ matches the latest web build.
• The Tauri custom title bar and CSP are part of the desktop security and window integration; update them using the project specialist guidance.

Docker Web App

docker build -t speechforge:web .
docker run --rm -p 3000:3000 speechforge:web

Open http://localhost:3000. Production serving must retain SPA fallback behavior.

Arch Linux Packaging

makepkg -si