Update app.py

app.py

@@ -1,101 +1,101 @@
-import os
-import requests
-import tempfile
-import gradio as gr
-from moviepy.editor import VideoFileClip
-from speechbrain.inference.interfaces import foreign_class
-import whisper
-from together import Together
-
-# Initialize Whisper once
-_whisper_model = whisper.load_model("base")
-
-# Initialize SpeechBrain classifier once
-_classifier = foreign_class(
-    source="warisqr7/accent-id-commonaccent_xlsr-en-english",
-    pymodule_file="custom_interface.py",
-    classname="CustomEncoderWav2vec2Classifier"
-)
-
-# Helper to download direct‐mp4 URL to a temp file
-def download_video(url: str) -> str:
-    resp = requests.get(url, stream=True)
-    resp.raise_for_status()
-    tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".mp4")
-    for chunk in resp.iter_content(8192):
-        tmp.write(chunk)
-    tmp.close()
-    return tmp.name
-
-# Helper to extract audio to a temp file
-def extract_audio(video_path: str) -> str:
-    tmp_audio = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3").name
-    clip = VideoFileClip(video_path)
-    clip.audio.write_audiofile(tmp_audio, logger=None)
-    clip.close()
-    return tmp_audio
-
-# Main pipeline
-def analyze_url(video_url):
-    try:
-        # 1. Download & extract
-        vid = download_video(video_url)
-        aud = extract_audio(vid)
-
-        # 2. Accent classification
-        out_prob, score, idx, lab = _classifier.classify_file(aud)
-        accent = lab[0]
-        conf_pct = round(float(score) * 100, 2)
-
-        # 3. Transcription
-        result = _whisper_model.transcribe(aud)
-        transcript = result["text"]
-
-        # 4. LLM analysis
-        api_key = "d2eac592fd335c7fd047814946f55e0c6fc26dbf75d88b0d9eb2be4a52108ea5"
-        client = Together(api_key=api_key)
-        prompt = f"""
-You are an English-speaking coach. Given this transcript of a spoken English audio with an {accent} accent and classification confidence {conf_pct}%:
-\"\"\"{transcript}\"\"\"
-
-Evaluate how confident the speaker sounds based on fluency, clarity, filler usage, professional English, and pacing.
-Provide:
-- A proficiency score between 0 and 100
-- A brief explanation
-"""
-        resp = client.chat.completions.create(
-            model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free",
-            messages=[{"role": "user", "content": prompt}]
-        )
-        analysis = resp.choices[0].message.content.strip()
-
-        # Clean up temp files
-        os.remove(vid)
-        os.remove(aud)
-
-        return accent, f"{conf_pct}%", transcript, analysis
-
-    except Exception as e:
-        return "Error", "", "", str(e)
-
-# Build Gradio interface
-with gr.Blocks(title="English Accent & Confidence Analyzer") as demo:
-    gr.Markdown("## 🎙️ English Accent Detection & Confidence Analysis")
-    with gr.Row():
-        inp = gr.Textbox(label="Direct MP4 Video URL", placeholder="https://...")
-        run = gr.Button("Analyze")
-    with gr.Row():
-        out1 = gr.Textbox(label="Detected Accent")
-        out2 = gr.Textbox(label="Accent Confidence")
-    out3 = gr.Textbox(label="Transcript", lines=5)
-    out4 = gr.Textbox(label="LLM Confidence Analysis", lines=10)
-
-    run.click(
-        fn=analyze_url,
-        inputs=inp,
-        outputs=[out1, out2, out3, out4],
-        api_name="analyze"
-    )
-
-if __name__ == "__main__":
-    demo.launch()
+import os
+import requests
+import tempfile
+import gradio as gr
+from moviepy import VideoFileClip
+from speechbrain.inference.interfaces import foreign_class
+import whisper
+from together import Together
+
+# Initialize Whisper once
+_whisper_model = whisper.load_model("base")
+
+# Initialize SpeechBrain classifier once
+_classifier = foreign_class(
+    source="warisqr7/accent-id-commonaccent_xlsr-en-english",
+    pymodule_file="custom_interface.py",
+    classname="CustomEncoderWav2vec2Classifier"
+)
+
+# Helper to download direct‐mp4 URL to a temp file
+def download_video(url: str) -> str:
+    resp = requests.get(url, stream=True)
+    resp.raise_for_status()
+    tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".mp4")
+    for chunk in resp.iter_content(8192):
+        tmp.write(chunk)
+    tmp.close()
+    return tmp.name
+
+# Helper to extract audio to a temp file
+def extract_audio(video_path: str) -> str:
+    tmp_audio = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3").name
+    clip = VideoFileClip(video_path)
+    clip.audio.write_audiofile(tmp_audio, logger=None)
+    clip.close()
+    return tmp_audio
+
+# Main pipeline
+def analyze_url(video_url):
+    try:
+        # 1. Download & extract
+        vid = download_video(video_url)
+        aud = extract_audio(vid)
+
+        # 2. Accent classification
+        out_prob, score, idx, lab = _classifier.classify_file(aud)
+        accent = lab[0]
+        conf_pct = round(float(score) * 100, 2)
+
+        # 3. Transcription
+        result = _whisper_model.transcribe(aud)
+        transcript = result["text"]
+
+        # 4. LLM analysis
+        api_key = "d2eac592fd335c7fd047814946f55e0c6fc26dbf75d88b0d9eb2be4a52108ea5"
+        client = Together(api_key=api_key)
+        prompt = f"""
+You are an English-speaking coach. Given this transcript of a spoken English audio with an {accent} accent and classification confidence {conf_pct}%:
+\"\"\"{transcript}\"\"\"
+
+Evaluate how confident the speaker sounds based on fluency, clarity, filler usage, professional English, and pacing.
+Provide:
+- A proficiency score between 0 and 100
+- A brief explanation
+"""
+        resp = client.chat.completions.create(
+            model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free",
+            messages=[{"role": "user", "content": prompt}]
+        )
+        analysis = resp.choices[0].message.content.strip()
+
+        # Clean up temp files
+        os.remove(vid)
+        os.remove(aud)
+
+        return accent, f"{conf_pct}%", transcript, analysis
+
+    except Exception as e:
+        return "Error", "", "", str(e)
+
+# Build Gradio interface
+with gr.Blocks(title="English Accent & Confidence Analyzer") as demo:
+    gr.Markdown("## 🎙️ English Accent Detection & Confidence Analysis")
+    with gr.Row():
+        inp = gr.Textbox(label="Direct MP4 Video URL", placeholder="https://...")
+        run = gr.Button("Analyze")
+    with gr.Row():
+        out1 = gr.Textbox(label="Detected Accent")
+        out2 = gr.Textbox(label="Accent Confidence")
+    out3 = gr.Textbox(label="Transcript", lines=5)
+    out4 = gr.Textbox(label="LLM Confidence Analysis", lines=10)
+
+    run.click(
+        fn=analyze_url,
+        inputs=inp,
+        outputs=[out1, out2, out3, out4],
+        api_name="analyze"
+    )
+
+if __name__ == "__main__":
+    demo.launch()