File size: 3,213 Bytes
4bd1d07
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
---
language: sv
license: mit
tags:
  - whisper
  - automatic-speech-recognition
  - sv
  - transformers.js
  - onnx
  - speech
  - audio
  - transcription
datasets:
  - common_voice
metrics:
  - wer
model-index:
  - name: whisper-base-onnx-web-v3
    results:
      - task:
          type: automatic-speech-recognition
        dataset:
          type: common_voice
          name: Common Voice Swedish
        metrics:
          - type: wer
            value: N/A
            name: Word Error Rate
---

# 🎤 whisper-base-onnx-web-v3

Fine-tuned Whisper model for Swedish transcription, optimized for web deployment with Transformers.js.

## 📋 Model Details

- **Base Model**: openai/whisper-base
- **Language**: Swedish (sv)
- **Task**: Speech Recognition / Transcription
- **Training Steps**: N/A
- **License**: MIT

## 🚀 Usage with Transformers.js

This model is optimized for browser-based transcription using Transformers.js:

```javascript
import { pipeline } from '@xenova/transformers';

// Load the model
const transcriber = await pipeline(
  'automatic-speech-recognition',
  'markusingvarsson/whisper-base-onnx-web-v3'
);

// Transcribe audio
const result = await transcriber(audioFile, {
  language: 'sv',
  task: 'transcribe',
  chunk_length_s: 30,
  stride_length_s: 5
});

console.log(result.text);
```

## 🐍 Usage with Python

```python
from transformers import pipeline

# Load pipeline
transcriber = pipeline(
    "automatic-speech-recognition",
    model="markusingvarsson/whisper-base-onnx-web-v3",
    device=0  # Use GPU if available
)

# Transcribe
result = transcriber(
    "audio.wav",
    generate_kwargs={"language": "sv", "task": "transcribe"}
)

print(result["text"])
```

## 📊 Performance

- **Word Error Rate (WER)**: N/A%
- **Model Size (ONNX)**: ~95MB (quantized)
- **Inference Speed**: 1-2x realtime on modern hardware

## 🎯 Intended Use

This model is designed for:
- Voice note transcription
- Meeting transcription
- Swedish podcast transcription
- Real-time speech-to-text in web browsers
- Accessibility applications

## 🔧 Training Details

- **Hardware**: GPU/CPU
- **Batch Size**: 8
- **Learning Rate**: 1e-5
- **Training Loss**: N/A

## 📁 Model Files

- `*.onnx`: ONNX model files for web deployment
- `config.json`: Model configuration
- `tokenizer.json`: Fast tokenizer for Transformers.js
- `processor_config.json`: Audio processing configuration

## 🌐 Demo

Try the model in your browser: [Coming Soon]

## 📝 Limitations

- Optimized for Swedish language only
- Best performance with clear audio (minimal background noise)
- May struggle with heavy dialects or very fast speech
- Maximum audio length: 30 seconds per chunk

## 🤝 Citation

If you use this model, please cite:

```bibtex
@misc{whisper_base_onnx_web_v3_2024,
  title={whisper-base-onnx-web-v3: Swedish Whisper for Web},
  author={markusingvarsson},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/markusingvarsson/whisper-base-onnx-web-v3}
}
```

## 🙏 Acknowledgments

- OpenAI for the original Whisper model
- Hugging Face for the tools and platform
- The Swedish NLP community

## 📄 License

This model is released under the MIT License.