Update app.py

list instead of queue for audio output chunk file mgt

app.py

@@ -11,7 +11,8 @@ from time import sleep
 import audioread
 import queue
 import threading
-from tempfile import NamedTemporaryFile
+from glob import glob
+import copy
 
 load_dotenv(override=True)
 key = os.getenv('OPENAI_API_KEY')
@@ -30,13 +31,10 @@ else:
     dp.mkdir(exist_ok=True)
     dataDir = '/data/'
 
-#speak_file = dataDir + "speek.wav"
+speak_file = dataDir + "speek.wav"
 
 client = OpenAI(api_key = key)
 
-qspeech = queue.Queue()
-qdelete = queue.Queue()
-
 #digits = ['zero: ','one: ','two: ','three: ','four: ','five: ','six: ','seven: ','eight: ','nine: ']
 
 abbrevs = {'St. ' : 'Saint ', 'Mr. ': 'mister ', 'Mrs. ':'mussus ', 'Mr. ':'mister ', 'Ms. ':'mizz '}
@@ -132,12 +130,9 @@ def genUsageStats(do_reset=False):
     result.append(['totals', f'{ttotal4mini_in}/{ttotal4mini_out}', f'{ttotal4o_in}/{ttotal4o_out}', f'audio:{totalAudio}',f'speech:{totalSpeech}'])
     return result       
 
-def clear():
-    while not qdelete.empty():
-        fname = qdelete.get()
-        if os.path.exists(fname):
-            os.remove(fname)
-    return [None, [], None]
+def new_conversation(user):
+    clean_up(user)
+    return [None, [], None, []]
 
 def updatePassword(txt):
     return [txt.lower().strip(), "*********"]
@@ -159,6 +154,11 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel):
         if prompt.startswith('gpt4'):
             gptModel = 'gpt-4o'
             prompt = prompt[5:]
+        if prompt.startswith("clean"):
+            user = prompt[6:]
+            response = f'cleaned all .wav files for {user}'
+            final_clean_up(user)
+            return [past, response, None, gptModel]
     if user_window in unames and pwd_window in pwdList:
         past.append({"role":"user", "content":prompt})
         completion = client.chat.completions.create(model=gptModel,
@@ -226,32 +226,68 @@ def pause_message():
 
 def set_speak_button(txt):
     vis = False
-    if len(txt) > 10:
+    if len(txt) > 2:
         vis = True
     return gr.Button(visible=vis)
 
-def clean_up():
-    while not qdelete.empty():
-        fname = qdelete.get()
-        if os.path.exists(fname):
-            os.remove(fname)
+def update_user(txt):
+    return txt.lower()
+
+# def clean_up(q):
+#     for fname in q:
+#         if os.path.exists(fname):
+#             os.remove(fname)
+#     return []
 
-def speech_worker(chunks=[]):
+def speech_worker(chunks=[],q=[]):
     for chunk in chunks:
-        response = client.audio.speech.create(model="tts-1", voice="fable", input=chunk, speed=0.85, response_format="wav")
-        tempFile = NamedTemporaryFile(mode='wb', delete=False)
-        tempFile.write(response.content)
-        tempFile.close()
-        qdelete.put (tempFile.name)
-        qspeech.put(tempFile.name)
+        fpath = q.pop(0)
+        response = client.audio.speech.create(model="tts-1", voice="fable", input=chunk, speed=0.85, response_format='wav')
+        with open(fpath, 'wb') as fp:
+            fp.write(response.content)
+
+# def gen_speech_file(user):
+#     global speech_file_index
+#     speech_file_index += 1
+#     fname = dataDir + f'user_speech{speech_file_index}.wav'
+#     return fname
+
+def gen_speech_file_names(user, cnt):
+    rv = []
+    for i in range(0, cnt):
+        rv.append(dataDir + f'{user}_speech{i}.wav')
+    return rv
 
+def final_clean_up(user):
+    flist = glob(dataDir + f'{user}_speech*.wav')
+    for fpath in flist:
+        try:
+            os.remove(fpath)
+        except:
+            continue
 
 with gr.Blocks() as demo:
+    history = gr.State([])
+    password = gr.State("")
+    user = gr.State("")
+    model = gr.State("gpt-4o-mini")
+    q = gr.State([])
+    qsave = gr.State([])
+
+    def clean_up(user):
+        flist = glob(dataDir + f'{user}_speech*.wav')
+        for fpath in flist:
+            try:
+                os.remove(fpath)
+            except:
+                continue
+
     def initial_audio_output(txt, user):
         global digits
         global abbrevs
-        while not qspeech.empty():
-            dud = qspeech.get()
+        q = []
+        if len(txt.strip()) < 5:
+            return ['None', q]
         for s,x in abbrevs.items():
             txt = txt.replace(s, x)
         words_in = txt.replace('**', '').splitlines(False)
@@ -310,25 +346,25 @@ with gr.Blocks() as demo:
         chunk = chunklist[0]
         if chunk.strip() == '':
             return gr.Audio(sources=None)
-        if len(chunklist) > 1:
-            threading.Thread(target=speech_worker, daemon=True, args=(chunklist[1:],)).start()
-        response = client.audio.speech.create(model="tts-1", voice="fable", input=chunk, speed=0.85, response_format="wav")
-        tempFile = NamedTemporaryFile(mode='wb', delete=False)
-        tempFile.write(response.content)
-        tempFile.close()
-        qdelete.put(tempFile.name)
-        return tempFile.name
+        fname_list = gen_speech_file_names(user, len(chunklist))
+        q = fname_list.copy()
+        qsave = fname_list.copy()
+        fname = q.pop(0)
+        if len(chunklist) > 0:
+            threading.Thread(target=speech_worker, daemon=True, args=(chunklist[1:],fname_list[1:])).start()
+        response = client.audio.speech.create(model="tts-1", voice="fable", input=chunk, speed=0.85, response_format='wav')
+        with open(fname, 'wb') as fp:
+            fp.write(response.content)
+        return [fname, q]
 
-    def gen_output_audio():
+    def gen_output_audio(q):
         try:
-            fname = qspeech.get(timeout=5)
+            fname = q.pop(0)
         except:
-            return gr.Audio(sources=None)
-        return fname
+            return [None, gr.Audio(sources=None)]
+        return [fname, q]
+
 
-    history = gr.State([])
-    password = gr.State("")
-    model = gr.State("gpt-4o-mini")
     gr.Markdown('# GPT Chat')
     gr.Markdown('Enter user name & password then enter prompt and click submit button.  Restart conversation if topic changes')
     gr.Markdown('You can enter prompts by voice.  Tap Record, speak, then tap Stop.' +
@@ -336,6 +372,7 @@ with gr.Blocks() as demo:
     # heading = gr.Label(value="GPT Chat", scale=2, color="Crimson" )
     with gr.Row():
         user_window = gr.Textbox(label = "User Name")
+        user_window.blur(update_user, user_window, user)
         pwd_window = gr.Textbox(label = "Password")
         pwd_window.blur(updatePassword, pwd_window, [password, pwd_window])
     with gr.Row():
@@ -350,16 +387,16 @@ with gr.Blocks() as demo:
         speak_output = gr.Button(value="Speak Dialog", visible=False)
     prompt_window = gr.Textbox(label = "Prompt or Question")
     output_window = gr.Textbox(label = "Dialog")
-    submit_button.click(chat, inputs=[prompt_window, user_window, password, history, output_window, model],
+    submit_button.click(chat, inputs=[prompt_window, user, password, history, output_window, model],
                          outputs=[history, output_window, prompt_window, model])
-    clear_button.click(clear, inputs=[], outputs=[prompt_window, history, output_window])
-    audio_widget.stop_recording(fn=transcribe, inputs=[user_window, password, audio_widget],
+    clear_button.click(fn=new_conversation, inputs=user, outputs=[prompt_window, history, output_window])
+    audio_widget.stop_recording(fn=transcribe, inputs=[user, password, audio_widget],
                                 outputs=[prompt_window])
     audio_widget.pause_recording(fn=pause_message, outputs=[prompt_window])
     reset_button.add(audio_widget)
     audio_out = gr.Audio(autoplay=True, visible=False)
-    audio_out.stop(fn=gen_output_audio, inputs=None, outputs = audio_out)
-    speak_output.click(fn=initial_audio_output, inputs=[output_window, user_window], outputs=audio_out)
+    audio_out.stop(fn=gen_output_audio, inputs=q, outputs = [audio_out, q])
+    speak_output.click(fn=initial_audio_output, inputs=[output_window, user], outputs=[audio_out, q])
     output_window.change(fn=set_speak_button, inputs=output_window,outputs=speak_output)
-    demo.unload(clean_up)
+    demo.unload(final_clean_up(user))
 demo.launch(share=True)