cleaning things up, fixing direct editing and bug catching

ai_tools.json

@@ -66,7 +66,7 @@
     "type": "function",
     "function": {
       "name": "edit_directly",
-      "description": "A function to be called when the user asks to directly edit either the lyrics or genre via the instrumental tags. Returns the current section or instrumental tags for the user to edit directly.",
+      "description": "A function to be called when the user asks to directly edit either the lyrics or genre instead of together via back-and-forth chat. Returns the current section or instrumental tags for the user to edit directly. Do not call if the user asks to revise together; instead call ask_question.",
       "parameters": {
         "type": "object",
         "properties": {

app.py

@@ -31,7 +31,7 @@ with gr.Blocks(css=css) as demo:
             gr.Markdown("""<center><font size=4>But first, let's generate a song seed to provide context to the AI Songwriter.</font></center>""")
             gr.Markdown("""<center><font size=3>If you're stuck thinking of a song idea, check out <a href="https://onestopforwriters.com/emotions" target="_blank">here</a>.</font></center>""")
             with gr.Row():
-                feeling_input = gr.Textbox(label="What do you want the song to capture? What's an emotion(s) that you've personally felt that should be in the song? More vulnerable, better the song.", placeholder='Enter your story', scale=2)
+                feeling_input = gr.Textbox(label="How are you feeling today?", placeholder='Enter your emotions', scale=2)
                 # audio_input = gr.Audio(sources=["upload"], type="numpy", label="Instrumental",
                 #                 interactive=True, elem_id="instrumental-input")
                 
@@ -69,7 +69,7 @@ with gr.Blocks(css=css) as demo:
             feeling_input.submit(generate_song_seed, inputs=[feeling_input], outputs=[instrumental_output]).then(clean_song_seed, inputs=[instrumental_output], outputs=[instrumental_output])
             
             def make_row_visible(x):
-                return gr.Row(visible=True), gr.Markdown("""<center><font size=4>Here it is! Hit 'Approve' to confirm this concept. Edit the concept directly or hit 'Try Again' to get another suggestion.</font></center>""", visible=True)
+                return gr.Row(visible=True), gr.Markdown("""<center><font size=4>Here it is! Hit 'Approve' to confirm this concept. Edit the concept directly or hit 'Try Again' to get another suggestion.</font></center>""", visible=False)
             def enable_button(x):
                 return gr.Button("Ready to Create", interactive=True)
             generate_seed_button.click(make_row_visible, inputs=[generate_seed_button], outputs=[concept_row, concept_desc])
@@ -101,7 +101,7 @@ with gr.Blocks(css=css) as demo:
 
             with gr.Row():
                 with gr.Column(scale=2):
-                    chatbot_history = gr.Chatbot(type="messages", value=starting_history, label='SongChat', placeholder=None, layout='bubble', bubble_full_width=False, height=500, show_copy_button=True)
+                    chatbot_history = gr.Chatbot(type="messages", value=starting_history, label='SongChat', placeholder=None, layout='bubble', bubble_full_width=False, height=500)
                     with gr.Row():
                         typical_responses = [textbox, submit]
                         
@@ -110,15 +110,15 @@ with gr.Blocks(css=css) as demo:
 
                         button_options = gr.State([])
                         button_dict = gr.State({
-                            "revise lyrics": "Can we revise the lyrics?",
-                            "re-revise lyrics": "Can we revise the lyrics?", 
+                            "revise lyrics": "Can we revise the lyrics together?",
+                            "re-revise lyrics": "Can we revise the lyrics together?", 
                             "edit lyrics directly": "Can I edit the lyrics directly for the whole section?",
                             "generate audio snippet": "Can you generate an audio snippet?", 
                             "continue revising" : "Can we continue revising this section?", 
                             "generate audio snippet with new lyrics": "Can you generate an audio snippet with these new lyrics?", 
                             "return to original instrumental": "Can you use the original clip for this section instead?", 
-                            "revise genre": "Can we revise the instrumental tags?",
-                            "re-revise genre": "Can we revise the instrumental tags?", 
+                            "revise genre": "Can we revise the instrumental tags together?",
+                            "re-revise genre": "Can we revise the instrumental tags together?", 
                             "revise genre directly": "Can I edit the genre directly for the whole song?",
                             "continue to next section": "Looks good! Let's move on to the next section.",
                             "merge snippets": "Can you merge this snippet into its full song?"
@@ -171,15 +171,15 @@ with gr.Blocks(css=css) as demo:
             continue_btn.click(get_starting_messages, inputs=[instrumental_textbox, title_input, blurb_input, genre_input, section_meanings], outputs=[messages, chatbot_history])
 
             with Modal(visible=False) as modal_0:
-                gr.Markdown("Welcome to the AI songwriter! The AI songwriter will help you write a song. You can chat with the AI, generate lyrics, and listen to audio snippets. Let's start by chatting with the AI.")
+                gr.Markdown("Welcome to the AI songwriter! The AI songwriter is a chatbot that will help you write a song. You can chat with the AI and guide it however you'd like. Let's start by chatting with the AI.")
             with Modal(visible=False) as modal:
-                gr.Markdown("The chatbot is run by an AI songwriter. It can respond to your conversations, generate lyrics and audio, and edit prior generations.\n\nNow, continue and respond to this second question from the AI songwriter.")
+                gr.Markdown("The AI songwriter can respond to your stories and requests, generate lyrics and audio, and edit prior generations.\n\nNow, continue and respond to this second question from the AI songwriter to get to know you.")
             with Modal(visible=False) as modal_1:
-                gr.Markdown("The AI songwriter has now proposed a first verse! You now have the option to hear an audio snippet, revise the lyrics, or continue to the next section. The latter two options continue the conversation, whereas the first starts audio generation models. Select the 'get audio snippet' button to continue to the next step.")
+                gr.Markdown("The AI songwriter has now proposed a first verse! After each generation from the AI, you'll receive a list of buttons to guide it further. Select the 'get audio snippet' button to continue to the next step.")
             with Modal(visible=False) as modal_2:
-                gr.Markdown("Awesome! You generated your first audio snippet./n/n As you work thru each section, the generated snippets are populated on the right panel. You'll be able to listen thru snippets as you work thru the song. \n\n "
-                            "The current section is also displayed on the right panel. You'll be able to revise sections via the chat or directly via the right panel. \n\n "
-                            "You're ready to start your official song! Hit the 'Start' button to start.")
+                gr.Markdown("Awesome! You generated your first audio snippet. The songwriter will continue for the each section for the rest of the song, revising and iterating with you. \n"
+                            "As the song gets generated, feel free to ask the songwriter any questions or guide it in any direction. \n"
+                            "You're ready to start your study with the AI Songwriter! Hit the 'Start' button to start.")
                 start_button = gr.Button("Start")
             
             continue_btn.click(lambda: Modal(visible=True), None, modal_0)
@@ -260,18 +260,24 @@ with gr.Blocks(css=css) as demo:
                 reset_button = gr.Button("Reset", scale=2)
             
                 def reset_chat(messages, chatbot_history):
-                    messages = messages[:2]
-                    chatbot_history = messages_to_history(messages[:2])
+                    messages = messages[:3]
+                    chatbot_history = messages_to_history(messages[:3])
                     return messages, chatbot_history, '', '', '', '', gr.HTML('<center>generating...</center>'), [], []
                 
-                reset_button.click(reset_chat, inputs=[messages, chatbot_history], outputs=[messages, chatbot_history, current_section, current_lyrics, curr_tags, clip_to_continue, curr_audio, generated_audios, button_options])
+                reset_button.click(reset_chat, inputs=[messages, chatbot_history], outputs=[messages, chatbot_history, current_section, current_lyrics, curr_tags, clip_to_continue, curr_audio, generated_audios, button_options]).then(
+                        update_response_options, [button_options, button_dict], typical_responses
+                )
             
 
             done.click(set_finish_query, inputs=[textbox], outputs=[textbox]).then(model_chat,
                 inputs=[genre_input, textbox, chatbot_history, messages, generated_audios],
-                outputs=[textbox, chatbot_history, messages, current_section, current_lyrics, curr_tags, clip_to_continue, curr_audio, generated_audios, button_options])
+                outputs=[textbox, chatbot_history, messages, current_section, current_lyrics, curr_tags, clip_to_continue, curr_audio, generated_audios, button_options]).then(
+                        update_response_options, [button_options, button_dict], typical_responses
+                )
 
-            demo.load(reset_chat, inputs=[messages, chatbot_history], outputs=[messages, chatbot_history, current_section, current_lyrics, curr_tags, clip_to_continue, curr_audio, generated_audios])
+            demo.load(reset_chat, inputs=[messages, chatbot_history], outputs=[messages, chatbot_history, current_section, current_lyrics, curr_tags, clip_to_continue, curr_audio, generated_audios, button_options]).then(
+                        update_response_options, [button_options, button_dict], typical_responses
+                )
             
             
             # with gr.Row():

chat.py

@@ -8,6 +8,8 @@ import regex as re
 from gradio_modal import Modal
 import gradio as gr
 import time
+from concurrent.futures import ThreadPoolExecutor
+
 
 # Load environment variables from .env file
 load_dotenv()
@@ -37,8 +39,14 @@ async def call_with_timeout(coro, timeout):
     except asyncio.TimeoutError:
         return "Timeout"
 
+executor = ThreadPoolExecutor()
+
+async def run_in_executor(func, *args):
+    loop = asyncio.get_event_loop()
+    return await loop.run_in_executor(executor, func, *args)
+
 
-def model_chat(genre_input, query: Optional[str], history: Optional[History], messages: Optional[Messages], generated_audios: List[Tuple[str, str, str]], auto=False) -> Tuple[str, History, Messages, str, str, str, str, str, List]:
+async def model_chat(genre_input, query: Optional[str], history: Optional[History], messages: Optional[Messages], generated_audios: List[Tuple[str, str, str]], auto=False) -> Tuple[str, History, Messages, str, str, str, str, str, List]:
     if query is None:
         query = ''
     with open('ai_tools.json') as f:
@@ -52,7 +60,7 @@ def model_chat(genre_input, query: Optional[str], history: Optional[History], me
         messages = messages + [{'role': 'user', 'content': query}]
 
 
-    messages_filtered = messages
+    messages_filtered = messages.copy()
     response_message = oai_client.chat.completions.create(
         model="gpt-4o",
         messages=messages_filtered,
@@ -302,7 +310,7 @@ def model_chat(genre_input, query: Optional[str], history: Optional[History], me
                     sections_list = re.findall(r'\[.*?\]', current_lyrics)
 
                     #current_lyrics = "\n".join(tool_query_args['sections_written'])
-                    song_link = make_song(current_lyrics, new_instrumental_tags)
+                    song_link = await call_with_timeout(run_in_executor(make_song, current_lyrics, new_instrumental_tags))
                     ## filter out suno link from tool query arg
                     while "https://audiopipe.suno.ai/?item_id=" not in song_link:
                         print("BUGGED OUT, trying again...")
@@ -314,7 +322,7 @@ def model_chat(genre_input, query: Optional[str], history: Optional[History], me
                             yield '', new_history, new_messages, '', '', '', None, None, generated_audios, []
                             return
                         time.sleep(5)
-                        song_link = make_song(current_lyrics, new_instrumental_tags)
+                        song_link = await call_with_timeout(run_in_executor(make_song, current_lyrics, new_instrumental_tags))
 
                     clip_id = song_link.split("https://audiopipe.suno.ai/?item_id=")[1]
 
@@ -350,7 +358,7 @@ def model_chat(genre_input, query: Optional[str], history: Optional[History], me
                     
                     new_instrumental_tags = songwriterAssistant.revise_instrumental_tags(snippet_instrumental_tags, user_instrumental_feedback)
 
-                    song_link = make_song(current_lyrics, new_instrumental_tags)
+                    song_link = await call_with_timeout(run_in_executor(make_song, current_lyrics, new_instrumental_tags))
                     ## filter out suno link from tool query arg
                     while "https://audiopipe.suno.ai/?item_id=" not in song_link:
                         print("BUGGED OUT, trying again...")
@@ -362,7 +370,7 @@ def model_chat(genre_input, query: Optional[str], history: Optional[History], me
                             yield '', new_history, new_messages, '', '', '', None, None, generated_audios, []
                             return
                         time.sleep(5)
-                        song_link = make_song(current_lyrics, new_instrumental_tags)
+                        song_link = await call_with_timeout(run_in_executor(make_song, current_lyrics, new_instrumental_tags))
                     clip_id = song_link.split("https://audiopipe.suno.ai/?item_id=")[1]
 
                     tool_message_instrumental = {'role': 'tool', 'tool_call_id': tool_call_id, 'name': tool_function_name, 'content': f'revised lyrics: {revised_lyrics}\nrevised instrumental tags: {new_instrumental_tags}, clip id: {clip_id}'}
@@ -381,7 +389,7 @@ def model_chat(genre_input, query: Optional[str], history: Optional[History], me
                     yield '', new_history, new_messages, tool_query_args["section_name"], revised_lyrics, new_instrumental_tags, clips_to_continue, f'<audio controls><source src="{song_link}" type="audio/mp3"></audio>', generated_audios, buttons
                 
                 elif tool_function_name == 'merge_all_snippets':
-                    updated_clip_url, updated_lyrics, updated_tags, clips_list = concat_snippets(tool_query_args['last_snippet_id'])
+                    updated_clip_url, updated_lyrics, updated_tags, clips_list = await call_with_timeout(run_in_executor(concat_snippets, tool_query_args['last_snippet_id']))
 
                     if updated_clip_url == "Timeout":
                         # Handle the timeout case
@@ -508,7 +516,7 @@ def model_chat(genre_input, query: Optional[str], history: Optional[History], me
                     snippet_instrumental_tags = tool_query_args['snippet_instrumental_tags']
 
                     snippet_clip_to_continue_from = tool_query_args.get('snippet_clip_to_continue_from', None)
-                    song_link = make_song(snippet_lyrics, snippet_instrumental_tags, snippet_clip_to_continue_from)
+                    song_link = await call_with_timeout(run_in_executor(make_song, snippet_lyrics, snippet_instrumental_tags, snippet_clip_to_continue_from), 45) 
 
                     if song_link == "Timeout":
                         tool_message = {
@@ -625,12 +633,13 @@ def model_chat(genre_input, query: Optional[str], history: Optional[History], me
                 else: 
                     print(f"Error: function {tool_function_name} does not exist")
         except Exception as e:
+            print(str(e))
             error_message = {
                 'role': 'assistant',
-                'content': f"An error occurred while processing your request: {str(e)}. Please re-phrase your request and try again."
+                'content': f"An error occurred while processing your request. Please re-phrase your request and try again."
             }
             messages_filtered.append(error_message)
-            yield '', messages_to_history(messages), messages_filtered, '', '', '', '', None, generated_audios, []
+            yield '', messages_to_history(messages_filtered), messages_filtered, '', '', '', '', None, generated_audios, []
 
             
     else: 

gpt_calls.py

@@ -64,7 +64,7 @@ class AI_Songwriter:
             instruction += "and complements the sections provided."
         else:
             instruction += "."
-        instruction += "You are also given a section description, genre, era, and overall description of the song."
+        instruction += "You are also given a section description, genre, era, and overall description of the song. Please do not literally quote and incorporate the ideas - be poetic and incorporate them artistically to express the emotion. "
 
         ## read in prompt lyrics from convo .txt and add it to instruction
         with open("prompts/write_section_ex.txt", "r") as f:
@@ -130,7 +130,7 @@ class AI_Songwriter:
         else:
             instruction += "."
         
-        instruction += "You are also given a genre, era, and the rest of the section."
+        instruction += "You are also given a genre, era, and the rest of the section. Please do not literally quote and incorporate the ideas - be poetic and incorporate them artistically to express the emotion. "
         
         with open("prompts/revise_section_ex.txt", "r") as f:
             convo = f.read()