Spaces:
Runtime error
Runtime error
File size: 6,650 Bytes
efca0f4 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 |
import tempfile
import os
import gc
from fastapi import FastAPI, File, UploadFile, HTTPException
from fastapi.responses import JSONResponse
from fastapi.middleware.cors import CORSMiddleware
import torch.nn.functional as F
import torchaudio
import torch
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
from src.transcription import SpeechEncoder
from src.sentiment import TextEncoder
from src.multimodal import MultimodalSentimentClassifier
# Configuration pour Hugging Face Spaces
HF_SPACE = os.getenv("HF_SPACE", "false").lower() == "true"
app = FastAPI(
title="API Multimodale de Transcription & Sentiment",
description="API pour l'analyse de sentiment audio en français",
version="1.0",
docs_url="/docs",
redoc_url="/redoc"
)
# Configuration CORS pour Hugging Face Spaces
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
# Précharge des modèles
print("Chargement des modèles pour l'API...")
try:
processor_ctc = Wav2Vec2Processor.from_pretrained(
"jonatasgrosman/wav2vec2-large-xlsr-53-french",
cache_dir="./models" if not HF_SPACE else None
)
model_ctc = Wav2Vec2ForCTC.from_pretrained(
"jonatasgrosman/wav2vec2-large-xlsr-53-french",
cache_dir="./models" if not HF_SPACE else None
)
speech_enc = SpeechEncoder()
text_enc = TextEncoder()
model_mm = MultimodalSentimentClassifier()
print("✅ Modèles chargés avec succès pour l'API")
except Exception as e:
print(f"❌ Erreur chargement modèles API: {e}")
raise
def transcribe_ctc(wav_path: str) -> str:
"""Transcription audio avec Wav2Vec2"""
try:
waveform, sr = torchaudio.load(wav_path)
if sr != 16000:
waveform = torchaudio.transforms.Resample(sr, 16000)(waveform)
if waveform.size(0) > 1:
waveform = waveform.mean(dim=0, keepdim=True)
inputs = processor_ctc(
waveform.squeeze().numpy(),
sampling_rate=16000,
return_tensors="pt",
padding=True
)
with torch.no_grad():
logits = model_ctc(**inputs).logits
pred_ids = torch.argmax(logits, dim=-1)
transcription = processor_ctc.batch_decode(pred_ids)[0].lower()
return transcription
except Exception as e:
raise HTTPException(status_code=500, detail=f"Erreur transcription: {str(e)}")
@app.get("/")
async def root():
"""Endpoint racine avec informations sur l'API"""
return {
"message": "API Multimodale de Transcription & Sentiment",
"version": "1.0",
"endpoints": {
"docs": "/docs",
"predict": "/predict",
"health": "/health"
},
"supported_formats": ["wav", "flac", "mp3"]
}
@app.get("/health")
async def health_check():
"""Vérification de l'état de l'API"""
return {
"status": "healthy",
"models_loaded": True,
"timestamp": "2024-01-01T00:00:00Z"
}
@app.post("/predict")
async def predict(file: UploadFile = File(...)):
"""
Analyse de sentiment audio
Args:
file: Fichier audio (WAV, FLAC, MP3)
Returns:
JSON avec transcription et sentiment
"""
# 1. Vérifier le type de fichier
if not file.filename or not file.filename.lower().endswith((".wav", ".flac", ".mp3")):
raise HTTPException(
status_code=400,
detail="Seuls les fichiers audio WAV/FLAC/MP3 sont acceptés."
)
# 2. Vérifier la taille du fichier (max 50MB)
content = await file.read()
if len(content) > 50 * 1024 * 1024: # 50MB
raise HTTPException(
status_code=400,
detail="Fichier trop volumineux. Taille maximale: 50MB"
)
# 3. Sauvegarder temporairement
suffix = os.path.splitext(file.filename)[1]
with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp:
tmp.write(content)
tmp_path = tmp.name
try:
# 4. Transcription
transcription = transcribe_ctc(tmp_path)
if not transcription.strip():
return JSONResponse({
"transcription": "",
"sentiment": {"négatif": 0.33, "neutre": 0.34, "positif": 0.33},
"warning": "Aucune transcription détectée"
})
# 5. Features multimodales
try:
audio_feat = speech_enc.extract_features(tmp_path)
text_feat = text_enc.extract_features([transcription])
# 6. Classification
logits = model_mm.classifier(torch.cat([audio_feat, text_feat], dim=1))
probs = F.softmax(logits, dim=1).squeeze().tolist()
labels = ["négatif", "neutre", "positif"]
sentiment = {labels[i]: round(probs[i], 3) for i in range(len(labels))}
except Exception as e:
# Fallback vers analyse textuelle uniquement
print(f"Erreur multimodal, fallback textuel: {e}")
sent_dict = TextEncoder.analyze_sentiment(transcription)
sentiment = {k: round(v, 3) for k, v in sent_dict.items()}
# 7. Nettoyage mémoire
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
return JSONResponse({
"transcription": transcription,
"sentiment": sentiment,
"filename": file.filename,
"file_size": len(content)
})
except Exception as e:
raise HTTPException(status_code=500, detail=f"Erreur lors de l'analyse: {str(e)}")
finally:
# 8. Nettoyage fichier temporaire
try:
os.remove(tmp_path)
except:
pass
@app.post("/predict_text")
async def predict_text(text: str):
"""
Analyse de sentiment textuel uniquement
Args:
text: Texte à analyser
Returns:
JSON avec sentiment
"""
try:
sent_dict = TextEncoder.analyze_sentiment(text)
sentiment = {k: round(v, 3) for k, v in sent_dict.items()}
return JSONResponse({
"text": text,
"sentiment": sentiment
})
except Exception as e:
raise HTTPException(status_code=500, detail=f"Erreur analyse textuelle: {str(e)}")
# Configuration pour Hugging Face Spaces
if __name__ == "__main__":
import uvicorn
uvicorn.run(
app,
host="0.0.0.0" if HF_SPACE else "127.0.0.1",
port=8000,
log_level="info"
) |