Skip navigation and go to main content

Transform Voice intoPrecision Content

Harness the power of elite ASR models to convert raw audio into structured, high-fidelity text with zero latency.

Read the guide
Animation: a sound wave flows into a pen nib and is written out as a transcript.

Try all six workflows

Pick a workflow on the left to see how it works in Scribis.

Scribis Studio
Auto-playing

Workflow

ASR ENGINE

Speech to Text

Accurate transcription for a range of audio formats.

01 / 06
TRANSCRIBING48 kHz / 32-bit float

INTERVIEW AUDIO00:00 / 00:42

Batch files, add punctuation, and apply your custom dictionary.

See Scribis in Action

Watch how Scribis transcribes high-fidelity audio with Parakeet & VibeVoice, performs text-driven video cuts, and exports dual subtitles in seconds.

  • VibeVoice & Parakeet Models
  • Text-Driven Rough Cuts
  • 100+ Languages & Dual Subs
Read User Documentation

Elite Audio-to-Text Model Library

A diverse collection of top-tier models optimized for various scenarios, from lightweight real-time recognition to ultra-precise multilingual transcription.

INDUSTRY STANDARD

Whisper Series

Benchmark multilingual ASR models by OpenAI, featuring superior noise resistance and exceptional accuracy.

Size 6 sizes, from 75 MB
PRO CHOICE

Parakeet tdt-v3

Optimized for European/American languages, supporting 25+ languages with high efficiency for long audio processing.

Size 652 MB
HIGH-SPEED

SenseVoice Small

An ultra-fast, low-latency model from Alibaba, supporting rich audio event detection.

Size 2 sizes, from 292 MB
LOCAL OPTIMIZED

Breeze ASR 25

Developed by MediaTek and NTU, optimized for localized scenarios with seamless Mandarin-English code-switching.

Size 3.09 GB
MULTILINGUAL

Qwen3 ASR

Open-source models from Alibaba Cloud supporting over 30 languages with elite multilingual and alignment capabilities.

Size 2 sizes, from 1.84 GB
LONG-FORM & DIARIZATION

VibeVoice ASR

Microsoft's elite 7B voice model, processing 60-minute audio in a single pass. Integrates ASR, speaker diarization, and timestamping into a structured log.

Size 4.81 GB(Q4_K)Parameters 7B
SPEECH-TO-MEANING

Voxtral Mini 3B

Mistral AI's lightweight 3B multimodal model, combining transcription and language understanding. Supports direct audio querying and function calling.

Size 2.65 GB(Q4_K)Parameters 3B
STREAMING REALTIME

Voxtral 4B Realtime

Mistral AI's native real-time model, delivering under 500ms transcription latency. Powered by causal audio encoding and adaptive latency control.

Size 2.4 GB(Q4_K)Parameters 4BLatency <500ms

Speech synthesis models

Voice design & cloning

Qwen3-TTS

Design a new voice from a text description or clone a speaker from reference audio, in 10 languages: Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish and Italian.

Sizes 0.6B / 1.7BLanguages 10
Languages & dialects

VoxCPM2

Covers 30 languages and 9 Chinese dialects, with voice design or cloning from a WAV reference, rendered at 48 kHz.

Size 2.83 GBLanguages 30Licence Apache 2.0

Works with the tools you already use

Beyond on-device models, connect cloud services with your own API keys. When a meeting starts, Scribis reminds you to record with a system notification.

AI services

Transcription, translation, language models and speech, with your own API key.

  • OpenAI
  • Google Gemini
  • OpenRouter
  • ElevenLabs
  • Groq
  • DeepSeek
  • Qwen
  • SiliconFlow
  • Azure
  • Doubao
  • Volcengine
  • DeepLX

Plus any OpenAI-compatible custom endpoint.

Meeting reminders

A native system notification appears when a meeting app starts using the microphone. Scribis never starts or stops recording on its own.

  • Zoom
  • Microsoft Teams
  • Webex
  • FaceTimemacOS
  • Slack
  • LINE
  • Discord
  • Google Meetbrowser

Web meetings are detected in Chrome, Safari, Edge, Arc, Firefox, Brave. Windows and macOS 14.2 or later; not available in the Mac App Store version.

Real workflows

From raw audio to ready-to-share

Scribis goes beyond transcripts: verify text against a script and on-screen captions, translate subtitles, synthesize speech from a speaker reference, and prepare the finished files for delivery.

Explore all features
LOCAL FIRST

Run transcription on your device

Choose an on-device speech model to process audio locally, with cloud services available when you need them.

PERSONALIZED CORRECTION

Your terminology, recognized correctly

Build replacement rules and pronunciation entries using IPA, Bopomofo, or Pinyin to improve recognition of specialist terms.

CAPTION EDITING

Edit video with your transcript

Align text to the waveform, rough-cut from the transcript, add zooms or mosaics, then export captions or a finished video.

VIDEO RESEARCH

Turn YouTube into a searchable library

Load a video and extract or translate its captions. With Gemini configured, ask questions, create timestamped summaries, and analyze selected scenes.

Every panel, one canvas

Assets and transcript, preview, scene style and the timeline all live in one editor window.

The Scribis video editor: assets and transcript on the left, the preview in the middle, scene and motion-graphic style settings on the right, and the storyboard and timeline below.

Precise alignment

Millisecond timecodes — drag any cue to match the audio.

Side-by-side bilingual

Original and translation side by side for bilingual subtitles.

Faster proofreading

Merge, split and batch-fix cues to proofread faster.

Live Intelligence.Instant Clarity.

Watch as your thoughts materialize. Our zero-delay pipeline renders text as the words leave your mouth, processed by our proprietary edge-compute lattice.

  • Ultra-fast audio processing pipeline
  • Automatic speaker identification
  • Real-time punctuation & formatting
Align38%

Subtitles burned into every frame

00:00:15:23
  1. Transcribe
  2. Align
  3. Burn-in
  4. Export

Ready to redefine your workflow?

Native Scribis apps for macOS and Windows are now available, bringing the ultimate AI voice experience to your desktop.

Talk to Sales

macOS (.dmg)

Installer from our website

Mac App Store

Apple Silicon only

Windows

Recommended: CPU is sufficient for Parakeet/SenseVoice transcription; others 4GB+ VRAM. 8GB+ VRAM (CUDA) recommended for translation.