DeLive

🎙️

Fourteen ASR Backends, One UI

Soniox, Volcengine, ElevenLabs, Mistral AI, Gladia, Deepgram, AssemblyAI, 60db, Cloudflare Workers AI, SiliconFlow, Groq, local OpenAI-compatible, local FunASR / SenseVoice, and local whisper.cpp — three execution modes in one app.

🧠

AI Review Desk

Full-page workspace with four tabs — Transcript (with AI side panel), Summary (overview, action items, keywords), multi-thread Chat with streaming output, and Markmap mind maps. AI Correction with smart text-source selection.

💬

Floating Caption Overlay

Always-on-top, draggable subtitle window with source, translated, and dual-line bilingual modes. Fully customizable font, color, shadow, and background.

🔒

Local-First & Secure

Sessions in IndexedDB, secrets in Electron safeStorage, context isolation, trusted-window IPC, CSP injection, and navigation guards. Your data never leaves your machine.

🌐

Open API & MCP Ecosystem

Local REST API (8 endpoints), real-time WebSocket streaming, standalone MCP server for Claude Desktop and Cursor, Agent Skill definition, and Agent Skills for one-call transcription inside any agent.

📁

File Transcription

Upload audio or video files and transcribe them with any of the fourteen providers, from cloud engines to fully local whisper.cpp and FunASR / SenseVoice. Speaker diarization, word-level timestamps, and language detection where the provider supports them.

🎨

Eight Themes, Light & Dark

Violet, Cyan, Rose, Green, Amber, Pink, Slate, and Orange accent palettes — each with full light and dark mode. New persistent sidebar navigation and Command Palette (Ctrl+K).

What's New

Coming in v2.3 (available now as a beta): two new providers, a Korean interface, and file transcription on every provider

📁

File Transcription on Every Provider

All fourteen providers can now transcribe uploaded audio or video, including fully local whisper.cpp (decoded to 16 kHz WAV and sent to whisper-server) and FunASR / SenseVoice. Providers no longer fall back to Soniox, and missing credentials show up as a failed job.

🔊

60db

New cloud provider with ~40 languages, including Indic languages mixed with English. Live capture runs through the embedded proxy; file jobs use the 60db REST API (up to 10 MB).

🖥️

FunASR / SenseVoice

New local provider backed by a self-hosted funasr-server. Choose SenseVoice (emotion and audio-event tags), Paraformer, or Fun-ASR-Nano, with no API cost.

🇰🇷

Korean Interface

Full Korean UI alongside Chinese and English, including tray menu, main-process strings, and localized error messages.

⬆️

Soniox V5

Soniox integration upgraded from V4 to V5; the real-time model defaults to stt-rt-v5.

⚡

ElevenLabs Turbo & Lite

Scribe v2 Realtime now offers Turbo and Lite variants next to the standard model.

🎙️

Gladia Solaria-3 for Files

File and async jobs can use Solaria-3; live streaming stays on Solaria-1.

🔌

Resilient Proxy Port

The embedded proxy tries ports 23456–23460 and then falls back to a free system port, and the main window opens even if the proxy cannot start.

🧪

373 Tests Passing

Test suite expanded to 373 tests across 44 files, covering the new providers, file transcription routing, and proxy handshakes.

DeLiveDesktop Transcription Workspace