Whisper Python サンプル：音声文字起こしの完全ガイド

• 音声テキスト化機能を開発するエンジニア
• オーディオデータを扱うデータサイエンティスト
• 実践的な Whisper Python サンプル を探している方

---

OpenAI Whisper とは？

• 99 以上の言語で音声を転写する
• 言語を自動検出する
• 音声を英語に翻訳する
• ノイズの多い音声やアクセントに対応する
• 長時間のオーディオを処理する

---

前提条件

• Python 3.8 以上
• パッケージマネージャー pip
• FFmpeg（オーディオ処理用）
• （任意）高速処理用の NVIDIA GPU

---

ステップ 1：Whisper をインストールする

pip install openai-whisper

FFmpeg をインストールする

brew install ffmpeg

sudo apt update
sudo apt install ffmpeg

---

ステップ 2：基本的な Whisper Python の例

import whisper

# Load the Whisper model
model = whisper.load_model("base")

# Transcribe audio file
result = model.transcribe("audio.mp3")

# Print the transcription
print(result["text"])

Hello everyone, welcome to today's meeting. We will discuss the project timeline and upcoming milestones.

---

ステップ 3：エラーハンドリング付きの完全な Python 例

import whisper
import os

def transcribe_audio(audio_path, model_size="base"):
    """
    Transcribe an audio file using Whisper.
    
    Args:
        audio_path (str): Path to the audio file
        model_size (str): Whisper model size (tiny, base, small, medium, large)
    
    Returns:
        dict: Transcription result with text and segments
    """
    try:
        # Check if audio file exists
        if not os.path.exists(audio_path):
            raise FileNotFoundError(f"Audio file not found: {audio_path}")
        
        # Load the Whisper model
        print(f"Loading Whisper model: {model_size}")
        model = whisper.load_model(model_size)
        
        # Transcribe the audio
        print(f"Transcribing: {audio_path}")
        result = model.transcribe(audio_path)
        
        return result
    
    except Exception as e:
        print(f"Error during transcription: {str(e)}")
        return None

# Example usage
if __name__ == "__main__":
    audio_file = "sample_audio.mp3"
    result = transcribe_audio(audio_file, model_size="base")
    
    if result:
        print("\nTranscription:")
        print(result["text"])

---

ステップ 4：言語検出を使った応用例

import whisper

model = whisper.load_model("base")

# Auto-detect language
result = model.transcribe("audio.mp3")
print(f"Detected language: {result['language']}")
print(f"Transcription: {result['text']}")

# Specify language explicitly
result_en = model.transcribe("audio.mp3", language="en")
result_zh = model.transcribe("audio.mp3", language="zh")

---

ステップ 5：タイムスタンプとセグメントを取得する

import whisper

model = whisper.load_model("base")
result = model.transcribe("audio.mp3")

# Print full transcription
print("Full Text:")
print(result["text"])

# Print segments with timestamps
print("\nSegments with Timestamps:")
for segment in result["segments"]:
    start = segment["start"]
    end = segment["end"]
    text = segment["text"]
    print(f"[{start:.2f}s - {end:.2f}s] {text}")

Full Text:
Hello everyone, welcome to today's meeting. We will discuss the project timeline.

Segments with Timestamps:
[0.00s - 2.50s] Hello everyone, welcome to today's meeting.
[2.50s - 5.80s] We will discuss the project timeline.

---

ステップ 6：オーディオを英語に翻訳する

import whisper

model = whisper.load_model("base")

# Translate to English
result = model.transcribe("spanish_audio.mp3", task="translate")

print("Translated text:")
print(result["text"])

---

ステップ 7：複数のオーディオファイルを処理する

import whisper
import os
from pathlib import Path

def batch_transcribe(audio_directory, model_size="base", output_dir="transcriptions"):
    """
    Transcribe all audio files in a directory.
    
    Args:
        audio_directory (str): Directory containing audio files
        model_size (str): Whisper model size
        output_dir (str): Directory to save transcriptions
    """
    # Create output directory
    os.makedirs(output_dir, exist_ok=True)
    
    # Load model once
    model = whisper.load_model(model_size)
    
    # Supported audio formats
    audio_extensions = ['.mp3', '.wav', '.m4a', '.flac', '.ogg']
    
    # Process each audio file
    audio_files = [
        f for f in os.listdir(audio_directory)
        if any(f.lower().endswith(ext) for ext in audio_extensions)
    ]
    
    for audio_file in audio_files:
        audio_path = os.path.join(audio_directory, audio_file)
        print(f"\nProcessing: {audio_file}")
        
        try:
            result = model.transcribe(audio_path)
            
            # Save transcription to file
            output_file = os.path.join(
                output_dir,
                Path(audio_file).stem + ".txt"
            )
            
            with open(output_file, "w", encoding="utf-8") as f:
                f.write(result["text"])
            
            print(f"✓ Saved: {output_file}")
            
        except Exception as e:
            print(f"✗ Error processing {audio_file}: {str(e)}")

# Example usage
batch_transcribe("audio_files/", model_size="base")

---

ステップ 8：SRT 字幕形式にエクスポートする

import whisper

def transcribe_to_srt(audio_path, output_path, model_size="base"):
    """
    Transcribe audio and save as SRT subtitle file.
    
    Args:
        audio_path (str): Path to audio file
        output_path (str): Path to save SRT file
        model_size (str): Whisper model size
    """
    model = whisper.load_model(model_size)
    result = model.transcribe(audio_path)
    
    # Generate SRT content
    srt_content = ""
    for i, segment in enumerate(result["segments"], start=1):
        start_time = format_timestamp(segment["start"])
        end_time = format_timestamp(segment["end"])
        text = segment["text"].strip()
        
        srt_content += f"{i}\n"
        srt_content += f"{start_time} --> {end_time}\n"
        srt_content += f"{text}\n\n"
    
    # Save SRT file
    with open(output_path, "w", encoding="utf-8") as f:
        f.write(srt_content)
    
    print(f"SRT file saved: {output_path}")

def format_timestamp(seconds):
    """Convert seconds to SRT timestamp format (HH:MM:SS,mmm)."""
    hours = int(seconds // 3600)
    minutes = int((seconds % 3600) // 60)
    secs = int(seconds % 60)
    millis = int((seconds % 1) * 1000)
    return f"{hours:02d}:{minutes:02d}:{secs:02d},{millis:03d}"

# Example usage
transcribe_to_srt("video.mp4", "subtitles.srt", model_size="base")

---

Whisper モデルサイズの比較

モデル	パラメータ	速度	精度	メモリ	用途
tiny	39M	⭐⭐⭐⭐⭐	⭐⭐	~1GB	高速テスト、シンプルな音声
base	74M	⭐⭐⭐⭐	⭐⭐⭐	~1GB	汎用
small	244M	⭐⭐⭐	⭐⭐⭐⭐	~2GB	バランス型
medium	769M	⭐⭐	⭐⭐⭐⭐⭐	~5GB	高精度が必要な場合
large	1550M	⭐	⭐⭐⭐⭐⭐⭐	~10GB	最高精度、ノイズ環境

---

Whisper Python のベストプラクティス

1. 適切なモデルサイズを選ぶ

# Fast and lightweight
model = whisper.load_model("tiny")  # Good for testing

# Balanced
model = whisper.load_model("base")  # Good for most cases

# High accuracy
model = whisper.load_model("medium")  # For important transcriptions

2. 長いオーディオを扱う

import whisper
from pydub import AudioSegment

def transcribe_long_audio(audio_path, chunk_length_ms=60000):
    """
    Transcribe long audio by splitting into chunks.
    
    Args:
        audio_path: Path to audio file
        chunk_length_ms: Length of each chunk in milliseconds
    """
    model = whisper.load_model("base")
    
    # Load audio
    audio = AudioSegment.from_file(audio_path)
    
    # Split into chunks
    chunks = []
    for i in range(0, len(audio), chunk_length_ms):
        chunks.append(audio[i:i + chunk_length_ms])
    
    # Transcribe each chunk
    full_text = []
    for i, chunk in enumerate(chunks):
        chunk_path = f"chunk_{i}.wav"
        chunk.export(chunk_path, format="wav")
        
        result = model.transcribe(chunk_path)
        full_text.append(result["text"])
        
        # Clean up chunk file
        os.remove(chunk_path)
    
    return " ".join(full_text)

3. 高速化のために GPU を使う

import whisper

# Whisper will automatically use GPU if available
model = whisper.load_model("base", device="cuda")

4. 精度向上のために言語を指定する

# If you know the language, specify it
result = model.transcribe("audio.mp3", language="en")

---

よくあるユースケース

ポッドキャストの転写

import whisper

model = whisper.load_model("medium")
result = model.transcribe("podcast_episode.mp3")

# Save transcript
with open("podcast_transcript.txt", "w") as f:
    f.write(result["text"])

会議メモ

import whisper
from datetime import datetime

model = whisper.load_model("base")
result = model.transcribe("meeting_recording.mp3")

# Create formatted meeting notes
notes = f"""
Meeting Notes - {datetime.now().strftime('%Y-%m-%d')}
========================================

{result["text"]}
"""

with open("meeting_notes.txt", "w") as f:
    f.write(notes)

動画字幕

import whisper

model = whisper.load_model("base")
result = model.transcribe("video.mp4")

# Generate VTT subtitle file
vtt_content = "WEBVTT\n\n"
for segment in result["segments"]:
    start = format_vtt_timestamp(segment["start"])
    end = format_vtt_timestamp(segment["end"])
    text = segment["text"].strip()
    vtt_content += f"{start} --> {end}\n{text}\n\n"

with open("subtitles.vtt", "w") as f:
    f.write(vtt_content)

---

よくあるトラブルと対処

問題 1：FFmpeg が見つからない

# Install FFmpeg
# macOS
brew install ffmpeg

# Ubuntu/Debian
sudo apt install ffmpeg

# Windows
# Download from ffmpeg.org and add to PATH

問題 2：メモリ不足

# Use a smaller model
model = whisper.load_model("tiny")  # Instead of "large"

# Or use CPU
model = whisper.load_model("base", device="cpu")

問題 3：処理が遅い

• より小さいモデル（tiny または base）を使う
• GPU アクセラレーションを有効にする
• オーディオをチャンクで処理する
• バッチジョブではマルチプロセスを使う

---

パフォーマンスのヒント

1. 可能なら GPU を使う — CPU より 10〜50 倍高速
2. タスクに合ったモデルサイズ — 単純な作業に「large」は不要
3. オーディオの前処理 — 無音除去、音量正規化
4. バッチ処理 — モデルは一度だけ読み込む
5. スレッド — I/O 中心の処理向け

---

Whisper Python と他のソリューション

項目	Whisper Python	Google Speech-to-Text	AssemblyAI
コスト	無料（ローカル）	従量課金	従量課金
オフライン	✅	❌	❌
精度	高い	高い	高い
セットアップ	中程度	簡単	簡単
長尺音声	✅	✅	✅
多言語	✅	✅	✅

---

完全例：本番向けスクリプト

#!/usr/bin/env python3
"""
Production-ready Whisper transcription script.
"""

import whisper
import argparse
import os
import json
from pathlib import Path
from datetime import datetime

def transcribe_file(
    audio_path,
    model_size="base",
    language=None,
    output_format="txt",
    output_dir=None
):
    """
    Transcribe an audio file with comprehensive output options.
    
    Args:
        audio_path: Path to audio file
        model_size: Whisper model size
        language: Language code (optional, auto-detected if None)
        output_format: Output format (txt, json, srt, vtt)
        output_dir: Output directory (default: same as audio file)
    """
    # Validate input file
    if not os.path.exists(audio_path):
        raise FileNotFoundError(f"Audio file not found: {audio_path}")
    
    # Set output directory
    if output_dir is None:
        output_dir = os.path.dirname(audio_path)
    os.makedirs(output_dir, exist_ok=True)
    
    # Load model
    print(f"Loading Whisper model: {model_size}")
    model = whisper.load_model(model_size)
    
    # Transcribe
    print(f"Transcribing: {audio_path}")
    transcribe_kwargs = {}
    if language:
        transcribe_kwargs["language"] = language
    
    result = model.transcribe(audio_path, **transcribe_kwargs)
    
    # Generate output filename
    base_name = Path(audio_path).stem
    output_path = os.path.join(output_dir, base_name)
    
    # Save based on format
    if output_format == "txt":
        with open(f"{output_path}.txt", "w", encoding="utf-8") as f:
            f.write(result["text"])
    
    elif output_format == "json":
        with open(f"{output_path}.json", "w", encoding="utf-8") as f:
            json.dump(result, f, indent=2, ensure_ascii=False)
    
    elif output_format == "srt":
        srt_content = generate_srt(result["segments"])
        with open(f"{output_path}.srt", "w", encoding="utf-8") as f:
            f.write(srt_content)
    
    elif output_format == "vtt":
        vtt_content = generate_vtt(result["segments"])
        with open(f"{output_path}.vtt", "w", encoding="utf-8") as f:
            f.write(vtt_content)
    
    print(f"✓ Transcription saved: {output_path}.{output_format}")
    print(f"  Language: {result['language']}")
    print(f"  Duration: {result['segments'][-1]['end']:.2f}s")
    
    return result

def generate_srt(segments):
    """Generate SRT subtitle content."""
    srt = ""
    for i, segment in enumerate(segments, start=1):
        start = format_timestamp(segment["start"])
        end = format_timestamp(segment["end"])
        text = segment["text"].strip()
        srt += f"{i}\n{start} --> {end}\n{text}\n\n"
    return srt

def generate_vtt(segments):
    """Generate VTT subtitle content."""
    vtt = "WEBVTT\n\n"
    for segment in segments:
        start = format_vtt_timestamp(segment["start"])
        end = format_vtt_timestamp(segment["end"])
        text = segment["text"].strip()
        vtt += f"{start} --> {end}\n{text}\n\n"
    return vtt

def format_timestamp(seconds):
    """Format timestamp for SRT."""
    hours = int(seconds // 3600)
    minutes = int((seconds % 3600) // 60)
    secs = int(seconds % 60)
    millis = int((seconds % 1) * 1000)
    return f"{hours:02d}:{minutes:02d}:{secs:02d},{millis:03d}"

def format_vtt_timestamp(seconds):
    """Format timestamp for VTT."""
    hours = int(seconds // 3600)
    minutes = int((seconds % 3600) // 60)
    secs = int(seconds % 60)
    millis = int((seconds % 1) * 1000)
    return f"{hours:02d}:{minutes:02d}:{secs:02d}.{millis:03d}"

def main():
    parser = argparse.ArgumentParser(
        description="Transcribe audio files using OpenAI Whisper"
    )
    parser.add_argument("audio", help="Path to audio file")
    parser.add_argument(
        "--model",
        default="base",
        choices=["tiny", "base", "small", "medium", "large"],
        help="Whisper model size"
    )
    parser.add_argument(
        "--language",
        default=None,
        help="Language code (e.g., 'en', 'zh', 'es')"
    )
    parser.add_argument(
        "--output-format",
        default="txt",
        choices=["txt", "json", "srt", "vtt"],
        help="Output format"
    )
    parser.add_argument(
        "--output-dir",
        default=None,
        help="Output directory"
    )
    
    args = parser.parse_args()
    
    transcribe_file(
        args.audio,
        model_size=args.model,
        language=args.language,
        output_format=args.output_format,
        output_dir=args.output_dir
    )

if __name__ == "__main__":
    main()

# Basic usage
python transcribe.py audio.mp3

# With options
python transcribe.py audio.mp3 --model medium --language en --output-format srt

# Save to specific directory
python transcribe.py audio.mp3 --output-dir ./transcriptions

---

まとめ

• Whisper は無料でオープンソース
• 99 以上の言語に対応
• オフラインで動作（API 呼び出し不要）
• 多くの用途で高精度
• Python プロジェクトへの組み込みが容易

---