Retranscription

Retranscription

Sometimes your first transcript isn't perfect. Maybe the wrong language was detected, you used a small model for speed during the meeting, or you want to try a cloud provider for better accuracy.

Think of it like re-developing a photo -the raw data (your audio recording) is still there. Retranscription sends it through a different processing pipeline to get a better result.

You'll find the retranscribe option in the Transcript tab of any meeting that has an audio recording.

Quick Start

1. Open any meeting from the sidebar
2. Switch to the Transcript tab
3. Click the Retranscribe button (refresh icon)
4. Choose a different provider, model, or language
5. Click Start Retranscription
6. Wait for processing to complete -your transcript is updated automatically

Transcript not good enough?
  → Open the meeting
  → Go to Transcript tab
  → Click Retranscribe
  → Pick a better provider/model/language
  → New transcript replaces the old one

When to Retranscribe

The Retranscribe Dialog

When you click the retranscribe button, a dialog opens with these options:

Provider Selection

Choose from all your configured transcription providers:

Model Selection

The model dropdown updates based on your chosen provider. For example:

• Whisper: tiny, base, small, medium, large-v3
• Groq: whisper-large-v3, whisper-large-v3-turbo

Language Selection

Set an explicit language hint or leave it on Auto-detect:

• Auto-detect: The engine guesses the spoken language (works well for single-language recordings)
• Specific language: Forces the engine to expect a particular language (best for non-English or multilingual recordings)

How Retranscription Works

When you start retranscription, Meetily processes the existing audio file through a simplified pipeline:

Stored Audio → Check Format → Transcribe → Save → Replace Original Transcript

Processing Stages

What you see during processing

• A progress bar shows the current percentage and stage name
• The dialog cannot be closed or dismissed while processing
• A Cancel button lets you abort at any time
• If cancelled, the original transcript remains unchanged

When it finishes

• A success toast shows the number of new transcript segments
• The Transcript tab updates to show the new transcription
• The original transcript is replaced -it cannot be recovered

Real-World Examples

Example 1: Fixing a wrong-language transcript

"I recorded a meeting in Japanese, but Whisper detected it as Chinese. Half the words are wrong."

Setup: Open the meeting → Transcript tab → Retranscribe → set Language to Japanese → Start Retranscription. The new transcript uses Japanese language models, producing a dramatically more accurate result.

Example 2: Upgrading from a quick model to a quality model

"During the live meeting I used tiny for real-time speed. Now I want the best possible transcript."

Setup: Open the meeting → Transcript tab → Retranscribe → select Whisper → choose large-v3 model → Start Retranscription. The larger model takes longer but catches nuances, technical terms, and quieter speakers that tiny missed.

Example 3: Switching from local to cloud for a noisy recording

"My local Whisper transcript is full of errors because the meeting had background construction noise."

Setup: Open the meeting → Transcript tab → Retranscribe → select Groq → choose whisper-large-v3 → Start Retranscription. Cloud providers often handle noisy audio better because they run on more powerful hardware with specialized noise-handling models.

Example 4: Re-processing an imported file

"I imported a recording and used the default settings. Now I want to try a specialized model on my self-hosted server."

Setup: Open the imported meeting → Transcript tab → Retranscribe → select OpenAI-Compatible → enter your server → select the specialized model → Start Retranscription. The self-hosted model processes the same audio with domain-specific accuracy.

Screenshots Guide

These are the key screens worth capturing for visual reference: