updating app

app.py

@@ -340,10 +340,9 @@ def streaming_process(audio_input, state):
         state.transcript_yo.append(yoruba)
 
     if audio_out is not None and len(audio_out) > 0:
-        # Convert to int16 PCM format for streaming Audio output
-        audio_out = np.clip(audio_out, -1.0, 1.0)
-        audio_int16 = (audio_out * 32767).astype(np.int16)
-        return (sr_out, audio_int16), format_live_log(state), state
+        # Ensure float32 in [-1, 1] range for autoplay Audio component
+        audio_out = np.clip(audio_out, -1.0, 1.0).astype(np.float32)
+        return (sr_out, audio_out), format_live_log(state), state
     else:
         return None, format_live_log(state), state
 
@@ -382,6 +381,8 @@ DESCRIPTION = """
 # Live Football Commentary \u2014 English \u2192 Yoruba
 
 Translate English football commentary into Yoruba speech in real-time.
+
+**Pipeline:** ASR (Whisper) \u2192 MT (NLLB-200) \u2192 TTS (MMS-TTS Yoruba)
 """
 
 STREAMING_INSTRUCTIONS = """
@@ -392,6 +393,7 @@ STREAMING_INSTRUCTIONS = """
 4. The transcript updates live below
 5. Click **Clear** to reset
 
+**Expected latency:** ~3\u20135 seconds behind your speech.
 """.format(chunk_dur=CHUNK_DURATION_S)
 
 EXAMPLES_TEXT = [
@@ -431,7 +433,6 @@ with gr.Blocks(
                         label="Yoruba Output",
                         type="numpy",
                         autoplay=True,
-                        streaming=True,
                     )
                     stream_log = gr.Markdown(
                         label="Live Transcript",
@@ -442,6 +443,8 @@ with gr.Blocks(
                 fn=streaming_process,
                 inputs=[stream_input, stream_state],
                 outputs=[stream_output, stream_log, stream_state],
+                time_limit=600,
+                stream_every=1.0,
             )
 
             clear_btn.click(