Update app.py

app.py

@@ -1,73 +1,59 @@
 import gradio as gr
+import tempfile
+import requests
 from moviepy.editor import VideoFileClip
-from speechbrain.pretrained import EncoderClassifier
 import torchaudio
-from pytube import YouTube
-import os
+import torch
+from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor, pipeline
+from transformers import WhisperProcessor, WhisperForConditionalGeneration
 
-CLASSIFIER = "Jzuluaga/accent-id-commonaccent_xlsr-en-english"
+# Load Models
+accent_model = Wav2Vec2ForSequenceClassification.from_pretrained("sreyan88/wav2vec2-large-xlsr-53-english-accent")
+feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("sreyan88/wav2vec2-large-xlsr-53-english-accent")
 
-def download_video(url):
-    """Handles YouTube and direct video links"""
-    if "youtube.com" in url:
-        yt = YouTube(url)
-        stream = yt.streams.filter(progressive=True, file_extension='mp4').first()
-        video_path = stream.download()
-        return video_path
-    else:  # Direct download (assumes URL is direct mp4 link)
-        # Download file locally
-        import requests
-        local_filename = "temp_video.mp4"
-        with requests.get(url, stream=True) as r:
-            r.raise_for_status()
-            with open(local_filename, 'wb') as f:
-                for chunk in r.iter_content(chunk_size=8192):
-                    f.write(chunk)
-        return local_filename
+whisper = pipeline("automatic-speech-recognition", model="openai/whisper-small")
 
-def extract_audio(video_path):
+# Helpers
+def download_and_extract_audio(url):
+    r = requests.get(url, stream=True)
+    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as video_file:
+        for chunk in r.iter_content(chunk_size=1024):
+            if chunk:
+                video_file.write(chunk)
+        video_path = video_file.name
     clip = VideoFileClip(video_path)
-    audio_path = "temp_audio.wav"
-    clip.audio.write_audiofile(audio_path, logger=None)
-    clip.close()
+    audio_path = video_path.replace(".mp4", ".wav")
+    clip.audio.write_audiofile(audio_path)
     return audio_path
 
 def classify_accent(audio_path):
-    classifier = EncoderClassifier.from_hparams(
-        source=CLASSIFIER,
-        savedir="pretrained_models/accent_classifier",
-        run_opts={"device":"cpu"}  # or "cuda" if GPU available
-    )
+    # Load and process audio
     waveform, sample_rate = torchaudio.load(audio_path)
-    prediction = classifier.classify_batch(waveform)
-    # prediction format: (scores, probabilities, embeddings, predicted_labels)
-    predicted_accent = prediction[3][0]
-    confidence = prediction[1].exp().max().item() * 100
-    return predicted_accent, f"{confidence:.2f}%"
+    inputs = feature_extractor(waveform[0], sampling_rate=sample_rate, return_tensors="pt", padding=True)
+    with torch.no_grad():
+        logits = accent_model(**inputs).logits
+    predicted_class = torch.argmax(logits, dim=1).item()
+    confidence = torch.softmax(logits, dim=1)[0, predicted_class].item()
+    labels = ["american", "australian", "british", "indian", "canadian"]  # match model classes
+    return labels[predicted_class], round(confidence * 100, 2)
 
-def process_video(url):
-    try:
-        video_path = download_video(url)
-        audio_path = extract_audio(video_path)
-        accent, confidence = classify_accent(audio_path)
-    finally:
-        # Cleanup temp files if they exist
-        for f in [video_path, audio_path]:
-            if os.path.exists(f):
-                os.remove(f)
-    return accent, confidence
+def full_pipeline(url):
+    audio_path = download_and_extract_audio(url)
+    accent, confidence = classify_accent(audio_path)
+    transcription = whisper(audio_path)["text"]
+    summary = f"The speaker’s accent is most likely **{accent.capitalize()}** with a confidence score of **{confidence}%**."
+    return transcription, accent.capitalize(), confidence, summary
 
-# Gradio interface
-iface = gr.Interface(
-    fn=process_video,
-    inputs=gr.Textbox(label="Enter Public Video URL (YouTube, Loom, direct MP4)"),
+# Gradio Interface
+gr.Interface(
+    fn=full_pipeline,
+    inputs=gr.Textbox(label="Public Video URL (MP4 or Loom)"),
     outputs=[
-        gr.Textbox(label="Detected Accent"),
-        gr.Textbox(label="Confidence Score")
+        gr.Textbox(label="Transcription"),
+        gr.Textbox(label="Accent"),
+        gr.Number(label="Confidence Score (%)"),
+        gr.Textbox(label="Summary"),
     ],
-    title="English Accent Classifier",
-    description="Paste a public video URL to detect the English accent and confidence score."
-)
-
-if __name__ == "__main__":
-    iface.launch()
+    title="Accent Classifier for English Speakers",
+    description="Paste a video URL to detect English accent and transcription. Built with Wav2Vec2 and Whisper on Hugging Face 🤗."
+).launch()