Spaces:
Sleeping
Sleeping
Code for real_time processing. For later testing
Browse files- views/real_time.py +80 -0
views/real_time.py
ADDED
|
@@ -0,0 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
################################
|
| 2 |
+
### NOT YET TESTED
|
| 3 |
+
###############################
|
| 4 |
+
|
| 5 |
+
import streamlit as st
|
| 6 |
+
import pyaudio
|
| 7 |
+
import wave
|
| 8 |
+
import torch
|
| 9 |
+
from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2Processor
|
| 10 |
+
import numpy as np
|
| 11 |
+
import time
|
| 12 |
+
|
| 13 |
+
# Charger le modèle Wav2Vec2 pour la classification des émotions
|
| 14 |
+
model_name = "superb/wav2vec2-base-superb-er" # Exemple de modèle pour la reconnaissance des émotions
|
| 15 |
+
processor = Wav2Vec2Processor.from_pretrained(model_name)
|
| 16 |
+
model = Wav2Vec2ForSequenceClassification.from_pretrained(model_name)
|
| 17 |
+
|
| 18 |
+
# Paramètres audio
|
| 19 |
+
CHUNK = 1024
|
| 20 |
+
FORMAT = pyaudio.paInt16
|
| 21 |
+
CHANNELS = 1
|
| 22 |
+
RATE = 16000
|
| 23 |
+
|
| 24 |
+
# Fonction pour prédire l'émotion à partir d'un segment audio
|
| 25 |
+
def predict_emotion(audio_data):
|
| 26 |
+
inputs = processor(audio_data, sampling_rate=RATE, return_tensors="pt", padding=True)
|
| 27 |
+
with torch.no_grad():
|
| 28 |
+
logits = model(**inputs).logits
|
| 29 |
+
predicted_id = torch.argmax(logits, dim=-1).item()
|
| 30 |
+
emotion = model.config.id2label[predicted_id]
|
| 31 |
+
return emotion
|
| 32 |
+
|
| 33 |
+
# Interface Streamlit
|
| 34 |
+
st.title("Détection des émotions en temps réel")
|
| 35 |
+
|
| 36 |
+
# Boutons pour démarrer et arrêter l'enregistrement
|
| 37 |
+
start_button = st.button("Démarrer l'enregistrement")
|
| 38 |
+
stop_button = st.button("Arrêter l'enregistrement")
|
| 39 |
+
|
| 40 |
+
# Zone de visualisation des émotions en temps réel
|
| 41 |
+
emotion_placeholder = st.empty()
|
| 42 |
+
final_emotion_placeholder = st.empty()
|
| 43 |
+
|
| 44 |
+
if start_button:
|
| 45 |
+
st.write("Enregistrement en cours...")
|
| 46 |
+
audio = pyaudio.PyAudio()
|
| 47 |
+
stream = audio.open(format=FORMAT, channels=CHANNELS, rate=RATE, input=True, frames_per_buffer=CHUNK)
|
| 48 |
+
|
| 49 |
+
frames = []
|
| 50 |
+
real_time_emotions = []
|
| 51 |
+
|
| 52 |
+
while not stop_button:
|
| 53 |
+
data = stream.read(CHUNK)
|
| 54 |
+
frames.append(data)
|
| 55 |
+
|
| 56 |
+
# Traitement en temps réel (par tranche de 1 seconde)
|
| 57 |
+
if len(frames) >= RATE // CHUNK:
|
| 58 |
+
audio_segment = np.frombuffer(b''.join(frames[-(RATE // CHUNK):]), dtype=np.int16)
|
| 59 |
+
emotion = predict_emotion(audio_segment)
|
| 60 |
+
real_time_emotions.append(emotion)
|
| 61 |
+
emotion_placeholder.line_chart(real_time_emotions) # Affichage graphique des émotions
|
| 62 |
+
|
| 63 |
+
# Arrêt de l'enregistrement
|
| 64 |
+
stream.stop_stream()
|
| 65 |
+
stream.close()
|
| 66 |
+
audio.terminate()
|
| 67 |
+
|
| 68 |
+
# Sauvegarde de l'audio enregistré
|
| 69 |
+
wf = wave.open("output.wav", "wb")
|
| 70 |
+
wf.setnchannels(CHANNELS)
|
| 71 |
+
wf.setsampwidth(audio.get_sample_size(FORMAT))
|
| 72 |
+
wf.setframerate(RATE)
|
| 73 |
+
wf.writeframes(b"".join(frames))
|
| 74 |
+
wf.close()
|
| 75 |
+
|
| 76 |
+
# Prédiction finale sur tout l'audio enregistré
|
| 77 |
+
full_audio_data = np.frombuffer(b''.join(frames), dtype=np.int16)
|
| 78 |
+
final_emotion = predict_emotion(full_audio_data)
|
| 79 |
+
|
| 80 |
+
final_emotion_placeholder.write(f"Émotion finale prédite : {final_emotion}")
|