Create app.py

app.py

@@ -0,0 +1,44 @@
+import gradio as gr
+from faster_whisper import WhisperModel
+import os
+
+# Load model on CPU (large-v2 multilingual)
+model = WhisperModel("large-v2", device="cpu", compute_type="int8")
+
+LANG_CODES = {
+    "Tamil": "ta",
+    "Malayalam": "ml"
+}
+
+def transcribe(audio, language_choice):
+    if audio is None:
+        return "No audio provided."
+
+    # Whisper transcription with minimal correction
+    segments, info = model.transcribe(
+        audio,
+        language=LANG_CODES[language_choice],
+        task="transcribe",
+        beam_size=1,
+        condition_on_previous_text=False,
+        initial_prompt=""
+    )
+
+    # Combine raw text from segments
+    full_text = "".join([seg.text for seg in segments])
+    return full_text.strip()
+
+# Gradio UI
+with gr.Blocks() as demo:
+    gr.Markdown("# 🎙 Whisper Large-v2 Raw Transcription\nMinimal correction, Tamil & Malayalam")
+
+    with gr.Row():
+        audio_input = gr.Audio(sources=["microphone", "upload"], type="filepath")
+        lang_choice = gr.Dropdown(choices=list(LANG_CODES.keys()), value="Tamil", label="Language")
+
+    output_text = gr.Textbox(label="Transcription")
+
+    submit_btn = gr.Button("Transcribe")
+    submit_btn.click(transcribe, inputs=[audio_input, lang_choice], outputs=output_text)
+
+demo.launch()