Spaces:
Running
Running
classificationV1
Browse files
app.py
CHANGED
@@ -1,7 +1,7 @@
|
|
1 |
import gradio as gr
|
2 |
import librosa
|
3 |
import torch
|
4 |
-
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, MarianMTModel, MarianTokenizer
|
5 |
|
6 |
# Charger le modèle de transcription pour le Darija
|
7 |
model = Wav2Vec2ForCTC.from_pretrained("boumehdi/wav2vec2-large-xlsr-moroccan-darija")
|
@@ -12,23 +12,39 @@ translation_model_name = "Helsinki-NLP/opus-mt-ar-en"
|
|
12 |
translation_model = MarianMTModel.from_pretrained(translation_model_name)
|
13 |
translation_tokenizer = MarianTokenizer.from_pretrained(translation_model_name)
|
14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
def transcribe_audio(audio):
|
16 |
-
"""
|
17 |
-
# Charger et prétraiter l'audio
|
18 |
audio_array, sr = librosa.load(audio, sr=16000)
|
19 |
input_values = processor(audio_array, return_tensors="pt", padding=True).input_values
|
20 |
|
21 |
-
# Obtenir les prédictions du modèle
|
22 |
logits = model(input_values).logits
|
23 |
tokens = torch.argmax(logits, axis=-1)
|
24 |
|
25 |
-
# Décoder la transcription en Darija
|
26 |
transcription = processor.decode(tokens[0])
|
27 |
-
|
28 |
-
# Traduire en anglais
|
29 |
translation = translate_text(transcription)
|
30 |
|
31 |
-
|
|
|
|
|
|
|
|
|
|
|
32 |
|
33 |
def translate_text(text):
|
34 |
"""Traduire le texte de l'arabe vers l'anglais"""
|
@@ -37,15 +53,31 @@ def translate_text(text):
|
|
37 |
translated_text = translation_tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
|
38 |
return translated_text
|
39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
# Interface utilisateur avec Gradio
|
41 |
with gr.Blocks() as demo:
|
42 |
-
gr.Markdown("# 🎙️ Speech-to-Text &
|
43 |
|
44 |
audio_input = gr.Audio(type="filepath", label="Upload Audio or Record")
|
45 |
-
submit_button = gr.Button("
|
|
|
46 |
transcription_output = gr.Textbox(label="Transcription (Darija)")
|
47 |
translation_output = gr.Textbox(label="Translation (English)")
|
|
|
|
|
48 |
|
49 |
-
submit_button.click(transcribe_audio,
|
|
|
|
|
50 |
|
51 |
demo.launch()
|
|
|
|
1 |
import gradio as gr
|
2 |
import librosa
|
3 |
import torch
|
4 |
+
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, MarianMTModel, MarianTokenizer, BertForSequenceClassification, AutoModel, AutoTokenizer
|
5 |
|
6 |
# Charger le modèle de transcription pour le Darija
|
7 |
model = Wav2Vec2ForCTC.from_pretrained("boumehdi/wav2vec2-large-xlsr-moroccan-darija")
|
|
|
12 |
translation_model = MarianMTModel.from_pretrained(translation_model_name)
|
13 |
translation_tokenizer = MarianTokenizer.from_pretrained(translation_model_name)
|
14 |
|
15 |
+
|
16 |
+
|
17 |
+
# Load AraBERT for Darija topic classification
|
18 |
+
arabert_model_name = "aubmindlab/bert-base-arabert"
|
19 |
+
arabert_tokenizer = AutoTokenizer.from_pretrained(arabert_model_name)
|
20 |
+
arabert_model = AutoModel.from_pretrained(arabert_model_name)
|
21 |
+
|
22 |
+
# Load BERT for English topic classification
|
23 |
+
bert_model_name = "bert-base-uncased"
|
24 |
+
bert_tokenizer = AutoTokenizer.from_pretrained(bert_model_name)
|
25 |
+
bert_model = BertForSequenceClassification.from_pretrained(bert_model_name, num_labels=3) # Adjust labels as needed
|
26 |
+
|
27 |
+
darija_topic_labels = ["Customer Service", "Retention Service", "Billing Issue"] # Adjust for Darija topics
|
28 |
+
english_topic_labels = ["Support Request", "Subscription Issue", "Payment Dispute"] # Adjust for English topics
|
29 |
+
|
30 |
+
|
31 |
def transcribe_audio(audio):
|
32 |
+
"""Convert audio to text, translate it, and classify topics in both Darija and English"""
|
|
|
33 |
audio_array, sr = librosa.load(audio, sr=16000)
|
34 |
input_values = processor(audio_array, return_tensors="pt", padding=True).input_values
|
35 |
|
|
|
36 |
logits = model(input_values).logits
|
37 |
tokens = torch.argmax(logits, axis=-1)
|
38 |
|
|
|
39 |
transcription = processor.decode(tokens[0])
|
|
|
|
|
40 |
translation = translate_text(transcription)
|
41 |
|
42 |
+
# Classify topics for both Darija and English
|
43 |
+
darija_topic = classify_topic(transcription, arabert_tokenizer, arabert_model, darija_topic_labels)
|
44 |
+
english_topic = classify_topic(translation, bert_tokenizer, bert_model, english_topic_labels)
|
45 |
+
|
46 |
+
return transcription, translation, darija_topic, english_topic
|
47 |
+
|
48 |
|
49 |
def translate_text(text):
|
50 |
"""Traduire le texte de l'arabe vers l'anglais"""
|
|
|
53 |
translated_text = translation_tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
|
54 |
return translated_text
|
55 |
|
56 |
+
def classify_topic(text, tokenizer, model, topic_labels):
|
57 |
+
"""Classify topic using BERT-based models"""
|
58 |
+
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)
|
59 |
+
with torch.no_grad():
|
60 |
+
outputs = model(**inputs)
|
61 |
+
predicted_class = torch.argmax(outputs.logits, dim=1).item()
|
62 |
+
|
63 |
+
return topic_labels[predicted_class] if predicted_class < len(topic_labels) else "Other"
|
64 |
+
|
65 |
+
|
66 |
# Interface utilisateur avec Gradio
|
67 |
with gr.Blocks() as demo:
|
68 |
+
gr.Markdown("# 🎙️ Speech-to-Text, Translation & Topic Classification")
|
69 |
|
70 |
audio_input = gr.Audio(type="filepath", label="Upload Audio or Record")
|
71 |
+
submit_button = gr.Button("Process")
|
72 |
+
|
73 |
transcription_output = gr.Textbox(label="Transcription (Darija)")
|
74 |
translation_output = gr.Textbox(label="Translation (English)")
|
75 |
+
darija_topic_output = gr.Textbox(label="Darija Topic Classification")
|
76 |
+
english_topic_output = gr.Textbox(label="English Topic Classification")
|
77 |
|
78 |
+
submit_button.click(transcribe_audio,
|
79 |
+
inputs=[audio_input],
|
80 |
+
outputs=[transcription_output, translation_output, darija_topic_output, english_topic_output])
|
81 |
|
82 |
demo.launch()
|
83 |
+
|