Update app.py

app.py

@@ -1,8 +1,8 @@
 import streamlit as st
 from transformers import pipeline
 import torch
-import librosa
 import tempfile
+from pydub import AudioSegment
 import math
 
 st.set_page_config(page_title="Urdu Speech-to-Text", layout="centered")
@@ -14,13 +14,13 @@ st.title("🎙️ Urdu Speech-to-Text (Whisper Turbo Urdu)")
 def get_device():
     if torch.cuda.is_available():
         st.success("GPU active ✓ (Fast Mode)")
-        return 0     # GPU
+        return 0
     else:
-        st.warning("GPU not available – switching to CPU (Slower but still works)")
-        return -1    # CPU fallback
+        st.warning("GPU not available – switching to CPU (slow mode)")
+        return -1
 
 # --------------------------
-# Load pipeline only once
+# Load ASR Model
 # --------------------------
 @st.cache_resource
 def load_asr(device):
@@ -35,42 +35,47 @@ def load_asr(device):
     )
 
 # --------------------------
-# Transcribe in chunks with progress bar
+# Chunk Transcription
 # --------------------------
 def transcribe_in_chunks(asr, audio_path):
-    y, sr = librosa.load(audio_path, sr=16000)
-    duration = librosa.get_duration(y=y, sr=sr)
+    audio = AudioSegment.from_file(audio_path)
 
-    chunk_sec = 30
-    total_chunks = math.ceil(duration / chunk_sec)
+    total_ms = len(audio)              # audio duration in milliseconds
+    chunk_ms = 30 * 1000               # 30 sec chunks
+    total_chunks = math.ceil(total_ms / chunk_ms)
 
-    st.info(f"⏳ Estimated time: {total_chunks} chunks")
+    st.info(f"⏳ Estimated chunks: {total_chunks}")
 
     progress = st.progress(0)
-    full_transcript = ""
+    full_text = ""
 
     for i in range(total_chunks):
-        start = i * chunk_sec
-        end = min((i + 1) * chunk_sec, duration)
+        start = i * chunk_ms
+        end = min((i+1) * chunk_ms, total_ms)
 
-        y_chunk = y[int(start * sr): int(end * sr)]
+        chunk = audio[start:end]
 
         with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
-            librosa.output.write_wav(tmp.name, y_chunk, sr)
+            chunk.export(tmp.name, format="wav")
             chunk_path = tmp.name
 
-        result = asr(chunk_path, return_timestamps=True)
+        result = asr(
+            chunk_path,
+            return_timestamps=True,
+            chunk_length_s=30,
+            stride_length_s=5
+        )
 
-        full_transcript += result["text"] + " "
+        full_text += result["text"] + " "
 
         progress.progress((i + 1) / total_chunks)
 
-    return full_transcript.strip()
+    return full_text.strip()
 
 # --------------------------
-# MAIN UI
+# APP UI
 # --------------------------
-uploaded_file = st.file_uploader("Upload audio", type=["mp3", "wav", "m4a", "ogg"])
+uploaded_file = st.file_uploader("Upload audio", type=["mp3","wav","m4a","ogg"])
 
 if uploaded_file:
     with tempfile.NamedTemporaryFile(delete=False, suffix=uploaded_file.name) as tmp:
@@ -79,11 +84,11 @@ if uploaded_file:
 
     st.success("✔ Audio uploaded")
 
-    # Choose device
     device = get_device()
     asr = load_asr(device)
 
-    st.info("Processing… please wait")
+    st.info("⏳ Transcribing audio…")
+
     transcript = transcribe_in_chunks(asr, audio_path)
 
     st.subheader("📝 Urdu Transcription")