Update app.py

app.py

@@ -21,9 +21,9 @@ def stop():
     global stop_processing
     stop_processing = True
 
-def transcribe_3600_seconds(audio_file):
+def transcribe_audio(audio_file):
     """
-    Process only the first 30 seconds of the audio, in small 5-second chunks.
+    Process up to 3600 seconds of the audio (1 hour) in 25-second chunks.
     Return partial text chunk by chunk (generator).
     """
     global stop_processing
@@ -32,22 +32,23 @@ def transcribe_3600_seconds(audio_file):
     # 1) Load at 16kHz
     waveform, sr = librosa.load(audio_file, sr=16000)
 
-    # 2) Truncate to the first 30 seconds
-    time_limit_s = 6000
+    # 2) Truncate to the first 3600 seconds (1 hour)
+    time_limit_s = 3600
     if len(waveform) > sr * time_limit_s:
         waveform = waveform[: sr * time_limit_s]
 
-    # Also limit if total is over 60 min (safety)
+    # Additional safety limit if total is somehow over 60 minutes
     max_audio_sec = 60 * 60
     if len(waveform) > sr * max_audio_sec:
         waveform = waveform[: sr * max_audio_sec]
 
-    # 3) Split that 30s portion into 5s chunks
+    # 3) Split that audio into 25-second chunks
     chunk_duration_s = 25
     chunk_size = sr * chunk_duration_s
     chunks = []
     for start_idx in range(0, len(waveform), chunk_size):
         chunk = waveform[start_idx : start_idx + chunk_size]
+        # Skip very small final pieces (less than 1 second)
         if len(chunk) < sr * 1:
             continue
         chunks.append(chunk)
@@ -74,7 +75,7 @@ def transcribe_3600_seconds(audio_file):
             predicted_ids = model.generate(
                 input_features,
                 attention_mask=attention_mask,
-                max_new_tokens=444,
+                max_new_tokens=444,   # keep under total token limit
                 do_sample=False,
                 forced_decoder_ids=forced_decoder_ids
             )
@@ -82,15 +83,22 @@ def transcribe_3600_seconds(audio_file):
         text_chunk = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
         partial_text += text_chunk + "\n"
 
+        # Stream updated partial text
         yield partial_text
 
-# Build Gradio UI with API support
-demo = gr.Interface(
-    fn=transcribe_3600_seconds,
-    inputs=gr.Audio(type="filepath"),
-    outputs="text",
-    title="Hebrew Whisper API",
-    api_name="transcribe"  # Enables API access
-)
+# Build Gradio UI
+with gr.Blocks() as demo:
+    gr.Markdown("## Hebrew Whisper (up to 1 hour, 25-second chunks)")
+
+    audio_input = gr.Audio(type="filepath", label="Upload Audio (Truncate to 1 hour)")
+    output_text = gr.Textbox(label="Partial Transcription")
+
+    start_btn = gr.Button("Start Transcription")
+    stop_btn = gr.Button("Stop Processing", variant="stop")
+
+    # Stream chunk-by-chunk without a progress bar
+    start_btn.click(transcribe_audio, inputs=audio_input, outputs=output_text)
+    stop_btn.click(stop)
 
-demo.launch()
+# Enable API by setting enable_api=True
+demo.launch(enable_api=True)