Stream audio output

infer-web.py

@@ -20,9 +20,9 @@ def build_standalone_ui():
 if __name__ == "__main__":
     app = build_merged_ui()
     if config.iscolab:
-        app.launch(share=True)
+        app.queue().launch(share=True)
     else:
-        app.launch(
+        app.queue().launch(
             server_name="localhost",
             inbrowser=not config.noautoopen,
             server_port=config.listen_port,

modules/merged_ui/main.py

@@ -17,7 +17,7 @@ def build_merged_ui():
         with gr.Tabs():
             with gr.TabItem("TTS-to-RVC Pipeline"):
                 gr.Markdown("### Generate speech with Spark TTS and convert with RVC")
-                gr.Markdown("*Note: For multi-sentence text, each sentence will be processed separately and then combined.*")
+                gr.Markdown("*Note: For multi-sentence text, each sentence will be processed separately and streamed as it’s ready.*")
                 
                 # TTS Generation Section
                 with gr.Row():
@@ -131,9 +131,9 @@ def build_merged_ui():
                 
                 with gr.Row():
                     vc_output1 = gr.Textbox(label="Output information", lines=10)
-                    vc_output2 = gr.Audio(label="Final concatenated audio")
+                    vc_output2 = gr.Audio(label="Streaming concatenated audio", autoplay=True)
                 
-                # Connect generate function to button
+                # Connect generate function to button with streaming enabled
                 generate_with_rvc_button.click(
                     generate_and_process_with_rvc,
                     inputs=[
@@ -152,7 +152,7 @@ def build_merged_ui():
                         rms_mix_rate0,
                         protect0,
                     ],
-                    outputs=[vc_output1, vc_output2],
+                    outputs=[vc_output1, vc_output2]
                 )
                 
                 # Connect modified_get_vc function for dropdown change

modules/merged_ui/utils.py

@@ -218,7 +218,7 @@ def generate_and_process_with_rvc(
     resample_sr, rms_mix_rate, protect
 ):
     """
-    Handle combined TTS and RVC processing for multiple sentences and save outputs to TEMP directories
+    Handle combined TTS and RVC processing for multiple sentences and yield outputs as they are processed.
     """
     # Ensure TEMP directories exist
     os.makedirs("./TEMP/spark", exist_ok=True)
@@ -227,7 +227,8 @@ def generate_and_process_with_rvc(
     # Split text into sentences
     sentences = split_into_sentences(text)
     if not sentences:
-        return "No valid text to process.", None
+        yield "No valid text to process.", None
+        return
     
     # Get next base fragment number
     base_fragment_num = 1
@@ -240,10 +241,12 @@ def generate_and_process_with_rvc(
     prompt_speech = prompt_wav_upload if prompt_wav_upload else prompt_wav_record
     prompt_text_clean = None if not prompt_text or len(prompt_text) < 2 else prompt_text
     
-    # Process each sentence
-    results = []
     info_messages = [f"Processing {len(sentences)} sentences..."]
+    results = []
     
+    # Yield initial message with no audio yet
+    yield "\n".join(info_messages), None
+
     for i, sentence in enumerate(sentences):
         spark_path, rvc_path, success, info = process_single_sentence(
             i, sentence, prompt_speech, prompt_text_clean,
@@ -256,22 +259,22 @@ def generate_and_process_with_rvc(
         info_messages.append(info)
         if success and rvc_path:
             results.append(rvc_path)
-    
+        
-    # If no sentences were successfully processed
+        # Build partial concatenation from results so far if any fragment exists
-    if not results:
+        if results:
-        return "\n".join(info_messages) + "\n\nNo sentences were successfully processed.", None
+            partial_output_path = f"./TEMP/partial_output_{base_fragment_num}.wav"
-    
+            concatenation_success = concatenate_audio_files(results, partial_output_path)
-    # Concatenate all successful RVC fragments
+            if not concatenation_success:
-    final_output_path = f"./TEMP/final_output_{base_fragment_num}.wav"
+                # Fallback: use the latest processed fragment
-    concatenation_success = concatenate_audio_files(results, final_output_path)
+                partial_output_path = results[-1]
-    
+        else:
-    if concatenation_success:
+            partial_output_path = None
-        info_messages.append(f"\nAll fragments concatenated successfully to: {final_output_path}")
+        
-        return "\n".join(info_messages), final_output_path
+        # Yield the current info and the partial audio so far
-    else:
+        yield "\n".join(info_messages), partial_output_path
-        # If concatenation failed but we have at least one successful fragment, return the first one
+
-        info_messages.append(f"\nFailed to concatenate fragments. Returning first successful fragment.")
+    # Optionally, yield one final update (could be identical to the last yield)
-        return "\n".join(info_messages), results[0]
+    yield "\n".join(info_messages), partial_output_path
 
 def modified_get_vc(sid0_value, protect0_value, file_index2_component):
     """