各種AIを使って動画ファイルから議事メモを作成する３

import asyncio
from audio_extractor import AudioExtractor
from audio_splitter import AudioSplitter
from summarizer_factory import SummarizerFactory
from speech_to_text import Transcriber


class MeetingMinutesCreationService():

  # 対象動画ファイル指定
  target_file = "./data/画面収録 2024-02-28 9.05.31.mov"

  async def execute(self):
    # 音声ファイル抽出
    audio_extractor = AudioExtractor(self.target_file)
    audio_file_name = audio_extractor.extract_audio()

    # 音声ファイルの分割
    splitter = AudioSplitter()
    separated_files: list[str] = splitter.split_wav(audio_file_name)

    # 音声→文字変換
    transcriber = Transcriber()
    transcribed_files = await transcriber.perform_transcription(separated_files)

    # 文字列の要約
    summarizer = SummarizerFactory.get_instance().create_summarizer_instance()
    summarizer.summarize_text(transcribed_files)

ibm-watson
moviepy
python-dotenv
scipy

openai==1.13.3  
langchain==0.1.10
langchain-openai

pip install -r requirements.txt

from enum import Enum
from typing import Dict
from summarizer import AbstractTextSummarizer
from text_summarizer_openai import OpenAITextSummarizer
from text_summarizer_watson import WatsonxTextSummarizer
from env_settings import EnvSettings

class LLMType(Enum):
  OPEN_AI='OPEN_AI'
  WATSONX_AI='WATSONX_AI'


class SummarizerFactory():
  __instance = None
  __env_settings = EnvSettings()
  __instance_dict: Dict[LLMType, AbstractTextSummarizer] = {
    LLMType.OPEN_AI: OpenAITextSummarizer(),
    LLMType.WATSONX_AI: WatsonxTextSummarizer(),
  }

  @classmethod
  def get_instance(cls):
    if not cls.__instance:
      cls.__instance = cls.__new__(cls)
    return cls.__instance

  def __init__(self) -> None:
    raise NotImplementedError('Cannot Generate Instance By Constructor Because This Class Is Singleton.')

  def create_summarizer_instance(self) -> AbstractTextSummarizer:
    llm_type = self.__env_settings.get_value_of("LLM_TYPE")
    return self.__instance_dict[LLMType[llm_type]]

from abc import ABCMeta, abstractmethod
import datetime
from env_settings import EnvSettings

class AbstractTextSummarizer(metaclass = ABCMeta):
  api_key: str
  env_settings = EnvSettings()
  output_path = f"./data/summary_result_{datetime.datetime.now().strftime('%y%m%d%H%M%S')}.txt"

  map_prompt_template = """以下の文章を要約して下さい。
  ------
  {text}
  ------
  """

  map_combine_template="""以下の文章を議事メモにまとめてください。決定事項、アクションアイテム、会議録の要約などです。
  ------
  {text}
  ------
  """

  @abstractmethod
  def __init__(self) -> None:
    pass

  @abstractmethod
  def summarize_text(self, target_files: list[str]) -> None:
    pass

from langchain_openai import ChatOpenAI
from langchain.prompts import PromptTemplate
from langchain.chains.summarize import load_summarize_chain
from langchain.docstore.document import Document

from summarizer import AbstractTextSummarizer

class OpenAITextSummarizer(AbstractTextSummarizer):

  def __init__(self) -> None:
    self.api_key = self.env_settings.get_value_of('OPEN_AI_API_KEY') # open ai API KEY    

  def summarize_text(self, target_files):
    if not target_files:
      return ""
    # プロンプトの定義
    map_first_prompt = PromptTemplate(template=self.map_prompt_template, input_variables=["text"])
    map_combine_prompt = PromptTemplate(template=self.map_combine_template, input_variables=["text"])
    # Map-Reduce方式
    map_chain = load_summarize_chain(
      llm=ChatOpenAI(temperature=0, model_name="gpt-3.5-turbo", api_key=self.api_key),
      reduce_llm=ChatOpenAI(temperature=0, model_name="gpt-4", api_key=self.api_key),
      collapse_llm=ChatOpenAI(temperature=0, model_name="gpt-4", api_key=self.api_key),
      chain_type="map_reduce",
      map_prompt=map_first_prompt,
      combine_prompt=map_combine_prompt,
      collapse_prompt=map_combine_prompt,
      token_max=5000,
      verbose=True
    )
    # API実行
    docs =[]
    for target_file in target_files:
      with open(target_file, "r") as file_obj:
        docs.append(Document(page_content=file_obj.read()))
    result = map_chain({"input_documents": docs}, return_only_outputs=True)
    # ファイル出力
    with open(self.output_path, "w") as file_obj:
      file_obj.write(result["output_text"])