whisper-transcription

1# Whisper Transcription
2 
3OpenAI Whisper provides accurate speech-to-text with word-level timestamps.
4 
5## Installation
6 
7```bash
8pip install openai-whisper
9```
10 
11## Model Selection
12 
13**Use the `tiny` model for fast transcription** - it's sufficient for most tasks and runs much faster:
14 
15| Model | Size | Speed | Accuracy |
16|-------|------|-------|----------|
17| tiny | 39 MB | Fastest | Good for clear speech |
18| base | 74 MB | Fast | Better accuracy |
19| small | 244 MB | Medium | High accuracy |
20 
21**Recommendation: Start with `tiny` - it handles clear interview/podcast audio well.**
22 
23## Basic Usage with Word Timestamps
24 
25```python
26import whisper
27import json
28 
29def transcribe_with_timestamps(audio_path, output_path):
30    """
31    Transcribe audio and get word-level timestamps.
32 
33    Args:
34        audio_path: Path to audio/video file
35        output_path: Path to save JSON output
36    """
37    # Use tiny model for speed
38    model = whisper.load_model("tiny")
39 
40    # Transcribe with word timestamps
41    result = model.transcribe(
42        audio_path,
43        word_timestamps=True,
44        language="en"  # Specify language for better accuracy
45    )
46 
47    # Extract words with timestamps
48    words = []
49    for segment in result["segments"]:
50        if "words" in segment:
51            for word_info in segment["words"]:
52                words.append({
53                    "word": word_info["word"].strip(),
54                    "start": word_info["start"],
55                    "end": word_info["end"]
56…