Spaces:
Sleeping
Sleeping
Upload 4 files
Browse files- Dockerfile +16 -0
- README.md +29 -0
- app.py +89 -0
- requirements.txt +8 -0
Dockerfile
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
FROM python:3.11-slim
|
| 2 |
+
|
| 3 |
+
WORKDIR /app
|
| 4 |
+
|
| 5 |
+
RUN apt-get update && apt-get install -y \
|
| 6 |
+
libsndfile1 ffmpeg \
|
| 7 |
+
&& rm -rf /var/lib/apt/lists/*
|
| 8 |
+
|
| 9 |
+
COPY requirements.txt .
|
| 10 |
+
RUN pip install --no-cache-dir -r requirements.txt
|
| 11 |
+
|
| 12 |
+
COPY app.py .
|
| 13 |
+
|
| 14 |
+
EXPOSE 7860
|
| 15 |
+
|
| 16 |
+
CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]
|
README.md
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
title: Deepfake Audio Detection API
|
| 3 |
+
emoji: 🎙️
|
| 4 |
+
colorFrom: purple
|
| 5 |
+
colorTo: blue
|
| 6 |
+
sdk: docker
|
| 7 |
+
app_port: 7860
|
| 8 |
+
pinned: false
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# Deepfake Audio Detection API
|
| 12 |
+
|
| 13 |
+
REST API for detecting AI-generated / deepfake audio using Wav2Vec2 (98.82% accuracy).
|
| 14 |
+
|
| 15 |
+
## Endpoints
|
| 16 |
+
|
| 17 |
+
**GET /health**
|
| 18 |
+
```json
|
| 19 |
+
{"status": "ok"}
|
| 20 |
+
```
|
| 21 |
+
|
| 22 |
+
**POST /detect** — send audio file as multipart form
|
| 23 |
+
```bash
|
| 24 |
+
curl -X POST https://rgriya-deepfake-audio-api.hf.space/detect \
|
| 25 |
+
-F "file=@your_audio.wav"
|
| 26 |
+
```
|
| 27 |
+
```json
|
| 28 |
+
{"label": "fake", "fake": 0.97, "real": 0.03}
|
| 29 |
+
```
|
app.py
ADDED
|
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Pure FastAPI — no Gradio dependency at all.
|
| 3 |
+
HuggingFace Spaces serves FastAPI apps directly.
|
| 4 |
+
|
| 5 |
+
GET /health → {"status":"ok"}
|
| 6 |
+
POST /detect → multipart file=<audio> → {"label":"fake","fake":0.97,"real":0.03}
|
| 7 |
+
"""
|
| 8 |
+
import io
|
| 9 |
+
import numpy as np
|
| 10 |
+
import soundfile as sf
|
| 11 |
+
from fastapi import FastAPI, File, UploadFile, HTTPException
|
| 12 |
+
from fastapi.responses import HTMLResponse
|
| 13 |
+
from transformers import pipeline
|
| 14 |
+
import torch
|
| 15 |
+
|
| 16 |
+
MODEL = "mo-thecreator/Deepfake-audio-detection"
|
| 17 |
+
print(f"Loading {MODEL} ...")
|
| 18 |
+
classifier = pipeline(
|
| 19 |
+
"audio-classification",
|
| 20 |
+
model=MODEL,
|
| 21 |
+
device=0 if torch.cuda.is_available() else -1,
|
| 22 |
+
)
|
| 23 |
+
print("Model ready.")
|
| 24 |
+
|
| 25 |
+
app = FastAPI(title="Deepfake Audio Detection")
|
| 26 |
+
|
| 27 |
+
|
| 28 |
+
def classify(raw_bytes: bytes) -> dict:
|
| 29 |
+
buf = io.BytesIO(raw_bytes)
|
| 30 |
+
try:
|
| 31 |
+
data, sr = sf.read(buf, dtype="float32", always_2d=False)
|
| 32 |
+
except Exception:
|
| 33 |
+
import librosa
|
| 34 |
+
buf.seek(0)
|
| 35 |
+
data, sr = librosa.load(buf, sr=16000, mono=True)
|
| 36 |
+
sr = 16000
|
| 37 |
+
|
| 38 |
+
if data.ndim > 1:
|
| 39 |
+
data = data.mean(axis=1)
|
| 40 |
+
peak = np.abs(data).max()
|
| 41 |
+
if peak > 1.0:
|
| 42 |
+
data /= peak
|
| 43 |
+
if sr != 16000:
|
| 44 |
+
import librosa
|
| 45 |
+
data = librosa.resample(data, orig_sr=sr, target_sr=16000)
|
| 46 |
+
|
| 47 |
+
out = io.BytesIO()
|
| 48 |
+
sf.write(out, data, 16000, format="WAV", subtype="PCM_16")
|
| 49 |
+
out.seek(0)
|
| 50 |
+
|
| 51 |
+
results = classifier(out.read())
|
| 52 |
+
print(f"Results: {results}")
|
| 53 |
+
scores = {r["label"]: float(r["score"]) for r in results}
|
| 54 |
+
return {"label": max(scores, key=scores.get), **scores}
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
@app.get("/", response_class=HTMLResponse)
|
| 58 |
+
def root():
|
| 59 |
+
return """
|
| 60 |
+
<html><body style="font-family:sans-serif;max-width:600px;margin:40px auto;padding:20px">
|
| 61 |
+
<h2>🎙️ Deepfake Audio Detection API</h2>
|
| 62 |
+
<p>Model: <code>mo-thecreator/Deepfake-audio-detection</code> (Wav2Vec2, 98.82% accuracy)</p>
|
| 63 |
+
<h3>Endpoints</h3>
|
| 64 |
+
<code>GET /health</code> — health check<br><br>
|
| 65 |
+
<code>POST /detect</code> — multipart <code>file=<audio></code><br>
|
| 66 |
+
<pre>{"label": "fake", "fake": 0.97, "real": 0.03}</pre>
|
| 67 |
+
<h3>Test</h3>
|
| 68 |
+
<form action="/detect" method="post" enctype="multipart/form-data">
|
| 69 |
+
<input type="file" name="file" accept="audio/*">
|
| 70 |
+
<button type="submit">Detect</button>
|
| 71 |
+
</form>
|
| 72 |
+
</body></html>
|
| 73 |
+
"""
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
@app.get("/health")
|
| 77 |
+
def health():
|
| 78 |
+
return {"status": "ok", "model": MODEL}
|
| 79 |
+
|
| 80 |
+
|
| 81 |
+
@app.post("/detect")
|
| 82 |
+
async def detect(file: UploadFile = File(...)):
|
| 83 |
+
try:
|
| 84 |
+
raw = await file.read()
|
| 85 |
+
print(f"/detect: {len(raw)} bytes filename={file.filename}")
|
| 86 |
+
return classify(raw)
|
| 87 |
+
except Exception as e:
|
| 88 |
+
import traceback; traceback.print_exc()
|
| 89 |
+
raise HTTPException(status_code=500, detail=str(e))
|
requirements.txt
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
fastapi>=0.110.0
|
| 2 |
+
uvicorn>=0.29.0
|
| 3 |
+
transformers>=4.39.0
|
| 4 |
+
torch>=2.0.0
|
| 5 |
+
torchaudio>=2.0.0
|
| 6 |
+
librosa>=0.10.0
|
| 7 |
+
soundfile>=0.12.0
|
| 8 |
+
python-multipart>=0.0.9
|