transcription.py

import os
from faster_whisper import WhisperModel
from pydub import AudioSegment
import string
import random
from datetime import datetime

# Matplotlibのキャッシュディレクトリを変更
os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib"

# Hugging Faceのキャッシュディレクトリを変更
os.environ["HF_HOME"] = "/tmp/huggingface"
os.environ["HUGGINGFACE_HUB_CACHE"] = "/tmp/huggingface"

class TranscriptionMaker():
    # 書き起こしファイルを吐き出すディレクトリを指定
    def __init__(self, output_dir="/tmp/data/transcriptions"):
        self.model = WhisperModel("base", device="cpu", download_root="/tmp/huggingface")
        self.output_dir = output_dir
        os.makedirs(self.output_dir, exist_ok=True)


    #音声ファイルのディレクトリを受け取り、書き起こしファイルを作成する
    def create_transcription(self,audio_directory):
        conversation = []

        #ディレクトリ内のファイルを全て取得
        if not os.path.isdir(audio_directory):
            raise ValueError(f"The specified path is not a valid directory: {audio_directory}")
        audio_files = self.sort_audio_files_in_directory(audio_directory)
        merged_segments = self.combine_audio(audio_files)
        merged_audio_directory = self.save_marged_segments(merged_segments, output_directory='/tmp/data/transcription_audio')
        merged_files = self.sort_audio_files_in_directory(merged_audio_directory)

        for audio_file in merged_files:
            if os.path.splitext(audio_file)[-1].lower() != '.wav':
                continue
            audio_path =  os.path.join(merged_audio_directory, audio_file)
            try:
                segments,info = list(self.model.transcribe(audio_path))
            except Exception as e:
                print(f"Error transcripting file {audio_path}: {e}")
                raise
            sorted_segments = sorted(segments, key=lambda s: s.start)
            results = []
            for segment in sorted_segments:
                results.append({
                    "start": segment.start,
                    "end": segment.end,
                    "text": segment.text
                })
            combined_text = "".join([result["text"] for result in results])
            speaker = os.path.basename(audio_file).split("_")[0]
            # 無音ならスキップ
            if not combined_text:
                continue
            conversation.append(f"{speaker}: {combined_text}<br>")
            
        #ファイルの書き込み。ファイル名は"transcription.txt"
        output_file=os.path.join(self.output_dir,"transcription.txt")
        print(conversation)
        try:
            with open(output_file,"w",encoding="utf-8") as f:
                for result in conversation:
                    f.write(result)
        except OSError as e:
            print(f"Error writing transcription file: {e}")
            raise
        return output_file
    
    # 受け取った音声ファイルを話者ごとに整理する
    def combine_audio(self,audio_files):
        if not audio_files:
            raise
        merged_segments = []
        current_speaker = None
        current_segment = [] 
        for segment in audio_files:
            speaker = os.path.basename(segment).split("_")[0]
            if speaker != current_speaker:
                # 話者が変わった場合はセグメントを保存
                if current_segment:
                    merged_segments.append((current_speaker, current_segment))
                current_speaker = speaker
                current_segment = [segment]
            else:
                # 話者が同一の場合はセグメントを結合
                current_segment.append(segment)
        # 最後のセグメントを保存
        if current_segment:
            merged_segments.append((current_speaker, current_segment))
        
        return merged_segments

    # ディレクトリ内の音声ファイルを並べ替える
    def sort_audio_files_in_directory(self, directory):
        files = os.listdir(directory)
        audio_files = [f for f in files if f.endswith(".wav")]

        audio_files.sort(key=lambda x: datetime.strptime(x.split("_")[1].split(".")[0], "%Y%m%d%H%M%S"))
        return [os.path.join(directory, f) for f in audio_files]
    
    def save_marged_segments(self,merged_segments,output_directory='/tmp/data/conversations'):
        if not merged_segments:
            print("merged_segmentsが見つかりませんでした。")
            raise
        
        conversation = []
        for speaker, segments in merged_segments:
            combined_audio = self.merge_segments(segments)
            conversation.append((speaker,combined_audio))
        if not os.path.exists(output_directory):
            os.makedirs(output_directory)

        for i, (speaker, combined_audio) in enumerate(conversation):
            current_time = datetime.now().strftime("%Y%m%d%H%M%S")
            filename = f"{speaker}_{current_time}.wav"
            file_path = os.path.join(output_directory,filename)
            combined_audio.export(file_path,format = "wav")
            print(f"Saved: {file_path}")

        return output_directory

    def merge_segments(self,segments):
        combined = AudioSegment.empty()  # 空のAudioSegmentを初期化
    
        for segment in segments:
            if isinstance(segment, str):
                # セグメントがファイルパスの場合、読み込む
                audio = AudioSegment.from_file(segment)
            elif isinstance(segment, AudioSegment):
                # セグメントがすでにAudioSegmentの場合、そのまま使用
                audio = segment
            else:
                raise ValueError("Invalid segment type. Must be file path or AudioSegment.")

            combined += audio  
        return combined 
    
    def generate_random_string(self,length):
        letters = string.ascii_letters + string.digits
        return ''.join(random.choice(letters) for i in range(length))

    def generate_filename(self,random_length):
        current_time = datetime.now().strftime("%Y%m%d%H%M%S")
        filename = f"{current_time}.wav"
        return filename