Update app.py

app.py

@@ -1,68 +1,71 @@
 import gradio as gr
 import whisperx
 import torch
-import tempfile
-import os
 
-# Load model once (important for speed)
+# Device setup
 device = "cuda" if torch.cuda.is_available() else "cpu"
 compute_type = "float16" if device == "cuda" else "int8"
 
-model = whisperx.load_model("small", device, compute_type=compute_type)
+# Load WhisperX model once
+model = whisperx.load_model("base", device, compute_type=compute_type)
 
 def transcribe(audio_file, language_code):
     if audio_file is None:
         return "Please upload audio"
 
-    # Save temp file
-    temp_audio = tempfile.NamedTemporaryFile(delete=False, suffix=".wav")
-    temp_audio.write(audio_file)
-    temp_audio.close()
+    try:
+        # Load audio directly from filepath
+        audio = whisperx.load_audio(audio_file)
 
-    # Load audio
-    audio = whisperx.load_audio(temp_audio.name)
+        # Transcribe (disable VAD for stability)
+        result = model.transcribe(
+            audio,
+            language=language_code,
+            vad_filter=False
+        )
 
-    # Transcribe
-    result = model.transcribe(audio, language=language_code)
+        # Load alignment model
+        model_a, metadata = whisperx.load_align_model(
+            language_code=result["language"],
+            device=device
+        )
 
-    # Align model for word timestamps
-    model_a, metadata = whisperx.load_align_model(
-        language_code=result["language"], device=device
-    )
+        # Align words
+        aligned_result = whisperx.align(
+            result["segments"],
+            model_a,
+            metadata,
+            audio,
+            device,
+            return_char_alignments=False
+        )
 
-    aligned_result = whisperx.align(
-        result["segments"],
-        model_a,
-        metadata,
-        audio,
-        device,
-        return_char_alignments=False
-    )
+        # Format output
+        output_lines = []
+        for seg in aligned_result["segments"]:
+            if "words" in seg:
+                for word in seg["words"]:
+                    start = round(word["start"], 2)
+                    end = round(word["end"], 2)
+                    text = word["word"]
+                    output_lines.append(f"[{start} - {end}] {text}")
 
-    # Format output
-    output = []
-    for seg in aligned_result["segments"]:
-        for word in seg["words"]:
-            start = round(word["start"], 2)
-            end = round(word["end"], 2)
-            text = word["word"]
-            output.append(f"[{start} - {end}] {text}")
+        return "\n".join(output_lines)
 
-    os.remove(temp_audio.name)
-
-    return "\n".join(output)
+    except Exception as e:
+        return f"Error: {str(e)}"
 
 
 # Gradio UI
 demo = gr.Interface(
     fn=transcribe,
     inputs=[
-        gr.Audio(type="binary", label="Upload Audio"),
-        gr.Textbox(label="Language Code (en, hi, etc.)", value="en"),
+        gr.Audio(type="filepath", label="Upload Audio"),
+        gr.Textbox(label="Language Code (en, hi, hi-IN, etc.)", value="en"),
     ],
     outputs=gr.Textbox(label="Word-level Transcription"),
     title="WhisperX Word-level Transcription",
-    description="Upload audio and get word-level timestamps"
+    description="Upload audio + language code → get word timestamps"
 )
 
 demo.launch()