Add Moroccan Darija extraction app2

app.py

@@ -2,7 +2,6 @@ import gradio as gr
 import torch
 from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC, pipeline
 import soundfile as sf
-import librosa
 
 # Load models
 # Transcription model for Moroccan Darija
@@ -10,46 +9,40 @@ processor = Wav2Vec2Processor.from_pretrained("boumehdi/wav2vec2-large-xlsr-moro
 transcription_model = Wav2Vec2ForCTC.from_pretrained("boumehdi/wav2vec2-large-xlsr-moroccan-darija")
 
 # Summarization model
-summarizer = pipeline("summarization", model="t5-small")
+summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
 
-# Function to resample audio to 16kHz if necessary
-def resample_audio(audio_path, target_sr=16000):
-    audio_input, original_sr = librosa.load(audio_path, sr=None)  # Load audio with original sampling rate
-    if original_sr != target_sr:
-        audio_input = librosa.resample(audio_input, orig_sr=original_sr, target_sr=target_sr)  # Resample to 16kHz
-    return audio_input, target_sr
-
-# Function to transcribe audio using Wav2Vec2
+# Function to transcribe audio
 def transcribe_audio(audio_path):
-    # Load and preprocess audio
-    audio_input, sample_rate = resample_audio(audio_path)
+    audio_input, sample_rate = sf.read(audio_path)
+    if sample_rate != 16000:
+        raise ValueError("Audio must be sampled at 16kHz.")
     inputs = processor(audio_input, sampling_rate=sample_rate, return_tensors="pt", padding=True)
-
-    # Get predictions
     with torch.no_grad():
         logits = transcription_model(**inputs).logits
-
-    # Decode predictions
     predicted_ids = torch.argmax(logits, dim=-1)
     transcription = processor.batch_decode(predicted_ids)[0]
     return transcription
 
+# Function to filter text by keywords
+def filter_text_by_keywords(text, keywords):
+    keyword_list = keywords.split(",")
+    filtered_sentences = [
+        sentence for sentence in text.split(". ") if any(keyword.strip().lower() in sentence.lower() for keyword in keyword_list)
+    ]
+    return ". ".join(filtered_sentences) if filtered_sentences else text
+
 # Function to transcribe and summarize
-def transcribe_and_summarize(audio_file):
-    # Transcription
+def transcribe_and_summarize(audio_file, keywords):
     transcription = transcribe_audio(audio_file)
-
-    # Check if transcription is long enough for summarization
-    if len(transcription.split()) < 10:  # Check if the transcription is too short for summarization
-        summary = "Transcription is too short for summarization."
-    else:
-        # Summarization
-        summary = summarizer(transcription, max_length=50, min_length=10, do_sample=False)[0]["summary_text"]
-        
+    filtered_text = filter_text_by_keywords(transcription, keywords)
+    summary = summarizer(filtered_text, max_length=100, min_length=30, do_sample=False)[0]["summary_text"]
     return transcription, summary
 
 # Gradio Interface
-inputs = gr.Audio(type="filepath", label="Upload your audio file")
+inputs = [
+    gr.Audio(type="filepath", label="Upload your audio file"),
+    gr.Textbox(label="Enter Keywords (comma-separated)", placeholder="e.g., customer, service, retention")
+]
 outputs = [
     gr.Textbox(label="Transcription"),
     gr.Textbox(label="Summary")
@@ -60,7 +53,10 @@ app = gr.Interface(
     inputs=inputs,
     outputs=outputs,
     title="Moroccan Darija Audio Transcription and Summarization",
-    description="Upload an audio file in Moroccan Darija to get its transcription and a summarized version of the content."
+    description=(
+        "Upload an audio file in Moroccan Darija to get its transcription and a summarized version. "
+        "Specify relevant keywords (comma-separated) to filter the transcription before summarization."
+    )
 )
 
 # Launch the app