ChatAIStreamer

Extention of ChatAIStream for voice generation.

The user of this library can

• pick up massegase from YouTube Chat and generate answer by ChatGPT.
• easily give role to ChatGPT.
• add callback of voice generation and get the return through answer_with_voice_cb.

How to install

Install from PyPI

• Install package to your environment.
  

  $ pip install chatai-streamer
  

Install from GitHub repository

• Clone this repository.
  

  $ clone https://github.com/GeneralYadoc/ChatAIStreamer.git
  

• Change directory to the root of the repository.
  

  $ cd ChatAIStreamer
  

• Install package to your environment.
  

  $ pip install .
  

How to use

• OpenAI API Key is necessary.

• You can use ChatAIStreamer class by implementing 'Add Implement' blocks of following codes.

  import sys
  import ChatAIStreamer as casr # Import this
  
  class myVoiceGenerator(casr.voiceGenerator):
    def generate(self, text):
      optimized_text = None
      voice = None
  
      #### Add Implementation here.
      # Optimize text for your TTS and assign to optimized_text variable.
      # Generate voice from optimized_text and assign to voice variable.
      # There is no restriction of the type of the voice.
      ####
  
      return optimized_text, voice
  
  # callback for getting answer of ChatGPT
  # The voice generated by voiceGenerator is given as an argument.
  def answer_with_voice_cb(user_message, completion, voice):
    #### Add Implementation here.
    # Implement for visualizing message (given as user_message).
    # Implement for visualizing ChatGPT answer (given as completion).
    # Implement for playing voice generated myVoiceGenerator.
    ####
  
  running = False
  
  # YouTube Video ID and ChatGPT API Key is given from command line in this example.
  if len(sys.argv) <= 2:
    exit(0)
  
  # Set params of getting messages from stream source.
  stream_params=casr.streamParams(video_id=sys.argv[1])
  
  # Set params of Chat AI.
  ai_params=casr.aiParams(
    api_key = sys.argv[2],
    system_role = "You are a cheerful assistant who speek English and can get conversation exciting with user."
  )
  
  # Set params of streamer
  streamer_params=casr.streamerParams(
    voice_generator=myVoiceGenerator(),
    answer_with_voice_cb=answer_with_voice_cb
  )
  
  # Create ChatAIStreamer instance.
  ai_streamer =casr.ChatAIStreamer(
    casr.params(
      stream_params=stream_params,
      ai_params=ai_params,
      streamer_params=streamer_params
    )
  )
  
  running = True
  
  # Wake up internal thread to get chat messages from stream and play voices of reading ChatGPT answers aloud.
  ai_streamer.start()
  
  # Wait any key inputted from keyboad.
  input()
  
  # Turn off runnging flag in order to finish playing voices of the sample.
  running=False
  
  # Finish generating voices.
  # Internal thread will stop soon.
  ai_streamer.disconnect()
  
  # terminating internal thread.
  ai_streamer.join()
  
  del ai_streamer
  
  

GttsAIStreamer - sample wrapper of ChatAIStreamer

• Sample class named GttsAIStreamer is provided with which you can play ChatGPT answers by gTTS voice.
• You can get gTTS voice and play it by the class.
• Sample codes using GttsAIStreamer exist here.

  # To execute this sample, please install ffmpeg.
  import sys
  import time
  import math
  import datetime
  import GttsAIStreamer as gasr
  
  # print sentence by a character incrementally.
  def print_incremental(st, interval_sec):
    for i in range(len(st)):
      if not running:
        break
      print(f"{st[i]}", end='')
      sys.stdout.flush()
      interruptible_sleep(interval_sec)
  
  # Customized sleep for making available of running flag interruption.
  def interruptible_sleep(time_sec):
    counter = math.floor(time_sec / 0.10)
    frac = time_sec - (counter * 0.10)
    for i in range(counter):
      if not running:
        break
      time.sleep(0.10)
    if not running:
      return
    time.sleep(frac)
  
  # callback for getting answer of ChatGPT
  # The voice generated by GttsGenerator is given.
  def answer_with_voice_cb(user_message, completion, voice):
    print(f"\n[{user_message.extern.author.name} {user_message.extern.datetime}] {user_message.message}\n")
    time_str = datetime.datetime.now().strftime ('%H:%M:%S')
    message = completion.choices[0]["message"]["content"]
  
    # Play the voice by VoicePlayer of GttsAIStreamer
    # If you use Python launcher named py, please create proper player as follows.
    # player = gasr.VoicePlayer(voice, python_command="py")
    player = gasr.VoicePlayer(voice)
    player.start()
  
    interruptible_sleep(3)
    print(f"[ChatGPT {time_str}] ", end='')
    print_incremental(message, 0.05)
    print("\n")
  
    # Wait finishing Playng the voice.
    player.join()
  
  running = False
  
  # YouTube Video ID and ChatGPT API Key is given from command line in this example.
  if len(sys.argv) <= 2:
    exit(0)
  
  # Set params of getting messages from stream source.
  stream_params=gasr.streamParams(video_id=sys.argv[1])
  
  # Set params of Chat AI.
  ai_params=gasr.aiParams(
    api_key = sys.argv[2],
    system_role = "You are a cheerful assistant who speek English and can get conversation exciting with user.",
  )
  
  # Create GttsVoiceGenerator
  # You can choose other language by its 'lang=' argument like 'ja'.
  # If you want to change language, Please change the assignment for system role of ai params also.
  # The instance isn't necessary for English generator.
  streamer_params=gasr.streamerParams(
    voice_generator=gasr.GttsGenerator(lang='en'),
    answer_with_voice_cb=answer_with_voice_cb
  )
  
  # Create GttsAIStreamer instance.
  # 'voice_generator=' is omittable for English generator.
  ai_streamer =gasr.GttsAIStreamer(
    gasr.params(
      stream_params=stream_params,
      ai_params=ai_params,
      streamer_params=streamer_params
    )
  )
  
  running = True
  
  # Wake up internal thread to get chat messages from stream and play gTTS voices of reading ChatGPT answers aloud.
  ai_streamer.start()
  
  # Wait any key inputted from keyboad.
  input()
  
  # Turn off runnging flag in order to finish playing gTTS voices of the sample.
  running=False
  
  # Finish generating gTTS voices.
  # Internal thread will stop soon.
  ai_streamer.disconnect()
  
  # terminating internal thread.
  ai_streamer.join()
  
  del ai_streamer

Usage of the sample

• Install ffmpeg
  For Linux: Execute following command.

  $ sudo apt-get install ffmpeg
  

  For Windows: Access here, download '*-win64-gpl.zip', extract the zip and move three exe files (ffmpeg.exe, ffprobe.exe, ffplay.exe) to the folder where you'll execute the sample or added path.
  
  For Mac: Access here, copy installation command to your terminal and push Enter key, and execute following command.

  brew install ffmpeg
  

• Execution command

  $ python3 ./sample.py VIDEO-ID OpenAI-API-KEY
  

• Output of the sample
  The outputs of the voices and the right window are provided by this sample.
  Left outputs are also available by ChatAIStreamer.

  

Arguments of Constructor

• ChatAIStreamer object can be configured with following params given to constructor.

  streamParams

  name	description	default
  video_id	String following after 'v=' in url of target YouTube live	-
  get_item_cb	Chat items are thrown to this callback	None
  pre_filter_cb	Filter set before internal queue	None
  post_filter_cb	Filter set between internal queue and get_item_cb	None
  max_queue_size	Max slots of internal queue (0 is no limit)	1000
  interval_sec	Polling interval of picking up items from YouTube	0.01 [sec]

  aiParams

  name	description	default
  api_key	API Key string of OpenAI	-
  system_role	ChatGPT role in convesation	"You are a helpful assistant."
  ask_cb	user message given to ChatGPT is thrown to this callback	None
  max_messages_in_context	Max messages in context given to ChatGPT	20
  answer_cb	ChatGPT answer is thrown to this callback	None
  max_queue_size	Max slots of internal queue (0 is no limit)	10
  model	Model of AI to be used.	None
  max_tokens_per_request	Max number of tokens which can be contained in a request	256
  interval_sec	Interval of ChatGPT API call	20.0 [sec]

  streamerParams

  name	description	default
  voice_generator	the instance of inherited class of voiceGenerater made by you	None
  answer_with_voice_cb	ChatGPT answer is thrown to this callback with voice data created by your get method of voiceGenerator	None
  max_queue_size	max size of internal queue which stocks GPT answer which will be given to voice Generator	1

Notice

• Please refer pytchat README to know the type of YouTube Chat item used by get_item_cb, pre_filter_cb and post filter_cb.
• Stamps in a message and messages consisted by stamps only are removed defaultly even if user doesn't set pre_filter_cb.
• Default value of interval_sec is 20.0, since free user of OpenAI API can get only 3 completions per minitue.
• The system role given by user remains ever as the oldest sentence of current context even if the number of messages is reached to the maximum, so ChatGPT doesn't forgot the role while current cunversation.

Methods

start()

• Start YouTube Chat polling, ChatGPT conversation and voice generation, then start calling user callbacks asyncronously.
• No arguments required, nothing returns.

join()

• Wait terminating internal threads kicked by start().
• No arguments required, nothing returns.

connect()

• Start YouTube Chat polling, ChatGPT conversation and voice generation, then start calling user callbacks syncronously.
• Lines following the call of the method never executen before terminate of internal threads.
• No arguments required, nothing returns.

disconnect()

• Request to terminate YouTube Chat polling, ChatGPT conversation, voice generation and calling user callbacks.
• Internal process will be terminated soon after.
• No arguments required, nothing returns.

And other threading.Thread public pethods are available.

Classes

voiceGenerator

• voiceGenerator class has hollowing method.

  name	description	default
  generate	virtual method to generate voice. String of text is given. Modified text and generated voice shoud be returended by this method. The type of modified text shoud be string. No restriction exists for voice type.	-

Callbacks

get_item_cb

• Callback for getting YouTube chat items.
• You can implement several processes in it.
• YouTube chat item is thrown as an argument.
• It's not be assumed that any values are returned.

pre_filter_cb

• pre putting queue filter.
• YouTube chat item is thrown as an argument.
• You can edit YouTube chat items before putting internal queue.
• If you want to get Complete items from YouTube, please implement this callback, since emoticons in a message and messages consisted by emoticons only are already removed from the items gotten in get_item_cb.
• It's required that edited chat item is returned.
• You can avoid putting internal queue by returning None.

post_filter_cb

• post getting queue filter
• You can edit YouTube chat items after popping internal queue.
• It's required that edited chat item is returned.
• You can avoid sending item to get_item_cb by returning None.

ask_cb

• Callback for getting questions that actually thrown to ChatGPT.
• If you register external info to user message when you put it, you can obtain the external info here.
• It's not be assumed that any values are returned.

answer_cb

• Callback for getting question and answer of ChatGPT.
• The type of completion is mentioned here.
• It's not be assumed that any values are returned.

answer_with_voice_cb

• Callback for getting question and answer of ChatGPT with voice data created by your generate method of voiceGenerator.
• The type of completion is mentioned here.
• if you modify the answer of ChatGPT on generate method of voiceGenerator, the modification will be reflect to completion argument of this callback.
• ask_cb and this callback are not always sequencial, since voice generation is performed on other thread in order to minimize latency.
• It's not be assumed that any values are returned.

Links

uses following libraries internally.

• chatai-stream
  Message broker between YouTube chat stream and ChatGPT.