Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,67 @@
|
|
| 1 |
-
---
|
| 2 |
-
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
title: Subtitle Generator
|
| 3 |
+
emoji: 🎤
|
| 4 |
+
colorFrom: blue
|
| 5 |
+
colorTo: green
|
| 6 |
+
sdk: gradio
|
| 7 |
+
sdk_version: 4.44.1
|
| 8 |
+
app_file: app.py
|
| 9 |
+
pinned: false
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# 🎤 Subtitle Generator
|
| 13 |
+
|
| 14 |
+
Automatically generate subtitles from audio/video files using OpenAI's Whisper model.
|
| 15 |
+
|
| 16 |
+
## Features
|
| 17 |
+
|
| 18 |
+
- 🎙️ **Speech-to-Text**: Transcribe audio using Whisper
|
| 19 |
+
- 🌍 **Auto Language Detection**: Automatically detects source language
|
| 20 |
+
- 📝 **Word-Level Timing**: Shows timing for each individual word
|
| 21 |
+
- 📄 **SRT Export**: Download subtitles in standard SRT format
|
| 22 |
+
- 🎬 **Video Support**: Extract audio from video files (MP4, MOV, MKV, AVI)
|
| 23 |
+
- 📖 **Story Mode**: Merge short sentences into longer segments
|
| 24 |
+
|
| 25 |
+
## How to Use
|
| 26 |
+
|
| 27 |
+
1. **Upload** an audio or video file
|
| 28 |
+
- Audio: MP3, WAV, OGG, M4A, FLAC, AAC
|
| 29 |
+
- Video: MP4, MOV, MKV, AVI
|
| 30 |
+
|
| 31 |
+
2. **Select** Whisper model size
|
| 32 |
+
- `tiny`: Fastest, least accurate
|
| 33 |
+
- `base`: Balanced (recommended)
|
| 34 |
+
- `small`: More accurate, slower
|
| 35 |
+
- `medium`: Most accurate, slowest
|
| 36 |
+
|
| 37 |
+
3. **Choose** processing mode
|
| 38 |
+
- **Podcast**: Keep original sentence structure
|
| 39 |
+
- **Story**: Merge short sentences (max 300 chars, 12 sec)
|
| 40 |
+
|
| 41 |
+
4. **Click** "Generate Subtitles" and wait for processing
|
| 42 |
+
|
| 43 |
+
5. **Download** the SRT file
|
| 44 |
+
|
| 45 |
+
## Model Information
|
| 46 |
+
|
| 47 |
+
| Model | Speed | Accuracy | Use Case |
|
| 48 |
+
|-------|-------|----------|----------|
|
| 49 |
+
| tiny | Fastest | Low | Quick previews |
|
| 50 |
+
| base | Fast | Good | General use (recommended) |
|
| 51 |
+
| small | Medium | Better | Podcasts, lectures |
|
| 52 |
+
| medium | Slow | Best | Professional transcription |
|
| 53 |
+
|
| 54 |
+
## Technical Details
|
| 55 |
+
|
| 56 |
+
- Uses **faster-whisper** for efficient CPU inference
|
| 57 |
+
- **VAD filter** removes silence/speech detection
|
| 58 |
+
- **Word timestamps** for precise alignment
|
| 59 |
+
- **Auto language detection** for multilingual support
|
| 60 |
+
- **Smart number formatting** (e.g., "1030" → "10:30" in time context)
|
| 61 |
+
|
| 62 |
+
## Notes
|
| 63 |
+
|
| 64 |
+
- First transcription will download the model (~150MB for base model)
|
| 65 |
+
- Processing time depends on file length and model size
|
| 66 |
+
- Videos are processed by extracting audio first
|
| 67 |
+
- All processing happens on CPU (no GPU required)
|