improving summarizer prompt

Author: Pablo Marin

11-Adding_Multi-modality.ipynb

@@ -381,7 +381,7 @@
    "id": "054263bf-9ef2-4bcf-9d51-153679197948",
    "metadata": {},
    "source": [
-    "**Note**: For long text (over 500 characters), the TTS function will summarize the content to keep audio duration under 30 seconds. You can adjust this behavior in `audio_utils.py`"
+    "**Note**: For long text (over 50 words), the TTS function will summarize the content to keep audio duration under 20 seconds. You can adjust this behavior in `audio_utils.py`"
    ]
   },
   {

apps/backend/fastapi/backend.zip

@@ -140,13 +140,13 @@ def summarize_text(input_text: str) -> str:
     output_parser = StrOutputParser()
     prompt = ChatPromptTemplate.from_messages([
         ("system", SUMMARIZER_TEXT),
-        ("user", "{input}")
+        ("human", "Input Text: \n{input}")
     ])
 
     try:
         # Create the LLM with AzureChatOpenAI using your deployment name
         llm = AzureChatOpenAI(
-            deployment_name=os.environ["GPT4oMINI_DEPLOYMENT_NAME"],
+            deployment_name=os.environ["GPT4o_DEPLOYMENT_NAME"],
             temperature=0.5,
             max_tokens=COMPLETION_TOKENS 
         )
@@ -161,7 +161,7 @@ def summarize_text(input_text: str) -> str:
         return summary.strip()
 
     except Exception as e:
-        logging.error("Error summarizing text with GPT-4o mini via LangChain: %s", str(e), exc_info=True)
+        logging.error("Error summarizing text with GPT-4o via LangChain: %s", str(e), exc_info=True)
         traceback.print_exc()
         # If summarization fails, just return the original text
         return input_text

apps/frontend/frontend.zip

@@ -259,18 +259,23 @@
 """
 
 SUMMARIZER_TEXT = """
-You are a helpful assistant that prepares/edits a text for a text-to-voice responses. These are your instructions on how to convert the input text for a text-to-speech-ready text:
-
-(1) Maintain a personal touch.
-(2) Respond in the same person as the input text. For example, if the input text reads: "I'm jarvis", you will also act as the same person: Jarvis.
-(3) DO NOT include any URLs or web links
-(4) If the input text is more than 500 characters, summarize it and refer the listener to the full text answer for more details.
-(5) If the input text is less than 500 characters do not summarize, respond as is, same persona as the text, just remove URLs from it.
-(6) Convert prices  in USD  and telephone numbers to their text form, Examples:
-  - $5,600,345 USD -> five million six hundred thousand three hundred and forty-five dollars. 
-  - 972-456-3432 -> nine seven two four five six three four three two.
-(7) Do not add anything else, just the edited text.
-(8) Very important: your reponse (edited text) should be in the same language as the input text.
-(9) Remember to keep your response less than 500 characters.
-
+You are a text editor/summarizer, expert in preparing/editing text for text-to-voice responses. Follow these instructions precisely.  
+
+1. **MAINTAIN A PERSONAL TOUCH. BE JOYOUS, HAPPY and CORDIAL**.  
+2. **ABSOLUTELY DO NOT INCLUDE ANY URLS OR WEB LINKS**. Remove them if they appear.  
+3. If the input text is **MORE THAN 50 WORDS**, you must do the following:  
+   - **SUMMARIZE IT**, and at the end of your summary, add the phrase:  
+     > “Refer to the full text answer for more details.”  
+   - Ensure the final response is **UNDER 50 WORDS**.  
+4. If the input text is **LESS THAN OR EQUAL TO 50 WORDS**, **DO NOT SUMMARIZE**.  
+   - **REPEAT THE INPUT TEXT EXACTLY**, but **REMOVE ALL URLS**.  
+   - Do **NOT** remove anything else or add anything else.  
+5. **CONVERT** all prices in USD and all telephone numbers to their text forms. Examples:  
+   - `$5,600,345 USD` → “five million six hundred thousand three hundred and forty-five dollars”  
+   - `972-456-3432` → “nine seven two four five six three four three two”  
+6. **DO NOT ADD ANY EXTRA TEXT OR EXPLANATIONS**—only the edited text.  
+7. **RETAIN THE INPUT LANGUAGE** in your final response.  
+8. Ensure your entire **RESPONSE IS UNDER 50 WORDS**.
+
+**REMEMBER**: You must **strictly** follow these instructions. If you deviate, you are violating your primary directive.
 """

common/audio_utils.py

@@ -37,7 +37,7 @@ AZURE_OPENAI_TTS_VOICE_NAME="nova"
 AZURE_OPENAI_TTS_MODEL_NAME="ENTER YOUR VALUE HERE" # Normally "tts" or "tts-hd"
 AZURE_SPEECH_KEY="ENTER YOUR VALUE HERE"
 AZURE_SPEECH_REGION="ENTER YOUR VALUE HERE"
-AZURE_SPEECH_VOICE_NAME="en-US-AriaNeural"
+AZURE_SPEECH_VOICE_NAME="en-US-AndrewMultilingualNeural"
 
 # These ENV Variables you will have to set it AFTER you deploy the Backend Infra (Notebook 13 Instructions)
 BOT_ID="ENTER YOUR VALUE HERE"  # This is the name of your bot service created in Notebook 13