Update app.py

app.py

@@ -45,26 +45,6 @@ llm = Llama.from_pretrained(
     verbose=False
 )
 
-def text_to_speech(text):
-    with tempfile.NamedTemporaryFile(delete=False, suffix='.wav') as temp_file:
-        with locker:
-            samples = tts.synthesize(text, "92")
-        output_file = temp_file.name
-        with wave.open(f"{output_file}", "w") as fp:
-            fp.setparams((1, 2, tts.get_sampling_rate(), len(samples), "NONE", "NONE"))
-            samples = np.ascontiguousarray(samples)
-            fp.writeframes(samples)
-    return output_file
-    
-
-def combine_audio_files(audio_files):
-    combined = AudioSegment.empty()
-    for audio_file in audio_files:
-        segment = AudioSegment.from_wav(audio_file)
-        combined += segment
-        os.remove(audio_file)  # Remove temporary files
-    return combined
-
 #guardrail model
 guard_llm = "llama-3.1-8b-instant"
 
@@ -119,14 +99,11 @@ async def greet(product,description):
     ]
     response = client.chat.completions.create(model=guard_llm, messages=messages, temperature=0)
     if response.choices[0].message.content != "not moderated":
-        audio_files = []
         a_list = ["Sorry, I can't proceed for generating marketing email. Your content needs to be moderated first. Thank you!"]
-        audio_file = text_to_speech(a_list[0])
-        audio_files.append(audio_file)
-        final_audio = combine_audio_files(audio_files)
-        yield (final_audio.frame_rate, final_audio)
+        with locker:
+            samples = tts.synthesize(a_list[0], "92")
+        yield (tts.get_sampling_rate(), samples)
     else:
-        audio_files = []
         output = llm.create_chat_completion(
         messages=[
             {
@@ -143,10 +120,9 @@ async def greet(product,description):
         for chunk in output:
             delta = chunk['choices'][0]['delta']
             if 'content' in delta:
-                audio_file = text_to_speech(delta.get('content', ''))
-                audio_files.append(audio_file)
-        final_audio = combine_audio_files(audio_files)
-        yield (final_audio.frame_rate, final_audio)
+                with locker:
+                    samples = tts.synthesize(delta.get('content', ''), "92")
+        yield (tts.get_sampling_rate(), samples)
 
 demo = gr.Interface(fn=greet, inputs=["text","text"], outputs=gr.Audio(), concurrency_limit=10)
 demo.launch()