Spaces:
Running
Running
| # | |
| # SPDX-FileCopyrightText: Hadad <[email protected]> | |
| # SPDX-License-Identifier: Apache-2.0 | |
| # | |
| import json # Import JSON module for encoding and decoding JSON data | |
| from src.tools.audio import AudioGeneration # Import AudioGeneration class to handle audio creation | |
| # Asynchronous handler for audio generation command | |
| async def audio_integration( | |
| input, # User input containing the /audio command and instruction | |
| new_history, # Conversation history in message format | |
| session_id, # Session ID for conversation context | |
| selected_model, # Selected AI model for generation | |
| jarvis, # AI backend function for generating responses | |
| mode, # Mode for AI response generation | |
| temperature, # Temperature parameter for AI | |
| top_k, # Top-k parameter for AI | |
| min_p, # Min-p parameter for AI | |
| top_p, # Top-p parameter for AI | |
| repetition_penalty # Repetition penalty for AI | |
| ): | |
| # Extract the audio instruction text after the '/audio' command prefix and strip whitespace | |
| audio_instruction = input[6:].strip() # Get instruction after /audio | |
| # If no instruction text is provided after the command, yield empty and exit early | |
| if not audio_instruction: # Check if instruction is empty | |
| yield [] # Yield empty list for missing instruction | |
| return # Exit function | |
| try: # Try block for audio generation | |
| # Asynchronously create audio content based on the instruction using AudioGeneration class | |
| audio = await AudioGeneration.create_audio(audio_instruction) # Generate audio | |
| # Serialize the audio data and instruction into a JSON formatted string for processing | |
| audio_generation_content = json.dumps({ | |
| "audio": audio, # Audio content or URL | |
| "audio_instruction": audio_instruction # Instruction for audio generation | |
| }) | |
| # Construct the conversation history including the audio generation result and formatting instructions | |
| audio_generation_result = ( | |
| new_history | |
| + [ | |
| { | |
| "role": "system", | |
| "content": ( | |
| "Audio generation result:\n\n" + audio_generation_content + "\n\n\n" | |
| "Show the audio using the following HTML audio tag format, where '{audio_link}' is the URL of the generated audio:\n\n" | |
| "<audio controls src='{audio_link}' style='width:100%; max-width:100%;'></audio>\n\n" | |
| "Please replace '{audio_link}' with the actual audio URL provided in the context.\n\n" | |
| "Then, describe the generated audio based on the above information.\n\n\n" | |
| "Use the same language as the previous user input or user request.\n" | |
| "For example, if the previous user input or user request is in Indonesian, explain in Indonesian.\n" | |
| "If it is in English, explain in English. This also applies to other languages.\n\n\n" | |
| ) | |
| } | |
| ] | |
| ) | |
| # Use async generator to get descriptive text about the generated audio from AI | |
| async for audio_description in jarvis( | |
| session_id=session_id, # Session ID | |
| model=selected_model, # Selected model | |
| history=audio_generation_result, # Updated history with audio result | |
| user_message=input, # User input | |
| mode=mode, # Mode for AI response | |
| temperature=temperature, # temperature parameter | |
| top_k=top_k, # top_k parameter | |
| min_p=min_p, # min_p parameter | |
| top_p=top_p, # top_p parameter | |
| repetition_penalty=repetition_penalty # repetition_penalty parameter | |
| ): | |
| yield [{"role": "tool", "content": audio_description}] # Yield audio description in tool role | |
| return # Exit after handling audio | |
| except Exception: # Exception handling for audio generation failure | |
| # If audio generation fails, let AI generate a contextual error message | |
| generation_failed = ( | |
| new_history | |
| + [ | |
| { | |
| "role": "system", | |
| "content": ( | |
| "Audio generation failed for the user's request. The user tried to generate audio with the instruction: '" | |
| + audio_instruction + "'\n\n\n" | |
| "Please explain to the user that audio generation failed and suggest they wait 15 seconds before trying again.\n" | |
| "Be helpful and empathetic in your response.\n\n\n" | |
| "Use the same language as the previous user input or user request.\n" | |
| "For example, if the previous user input or user request is in Indonesian, explain in Indonesian.\n" | |
| "If it is in English, explain in English. This also applies to other languages.\n\n\n" | |
| ) | |
| } | |
| ] | |
| ) | |
| # Use AI to generate a contextual error message | |
| async for error_response in jarvis( | |
| session_id=session_id, # Session ID | |
| model=selected_model, # Selected model | |
| history=generation_failed, # History with error context | |
| user_message=input, # User input | |
| mode="/no_think", # Use non-reasoning mode for error handling | |
| temperature=0.7, # Fixed temperature for more consistent error messages | |
| top_k=20, # Limit token sampling | |
| min_p=0, # Minimum probability threshold | |
| top_p=0.8, # Nucleus sampling threshold | |
| repetition_penalty=1 # No repetition penalty | |
| ): | |
| yield [{"role": "tool", "content": error_response}] # Yield error response in tool role | |
| return # Exit after error handling |