Configurazione Whisper con Docker: guida completa per eseguire OpenAI Whisper in Docker

Introduzione

• Distribuire ovunque – lo stesso container su qualsiasi piattaforma compatibile con Docker
• Isolare le dipendenze – evitare conflitti con i pacchetti di sistema
• Scalare facilmente – avviare più container per elaborazione parallela
• Controllare le versioni – fissare versioni e configurazioni specifiche di Whisper
• Semplificare il deployment – un solo comando per tutto

---

Perché usare Docker per Whisper?

Vantaggi della containerizzazione

• stesso ambiente in sviluppo, staging e produzione
• nessun conflitto di dipendenze
• build riproducibili

• esecuzione su qualsiasi piattaforma con Docker
• migrazione semplice tra server
• deployment indipendente dal cloud

• nessuna interferenza con il sistema host
• disinstallazione pulita (rimuovere il container)
• sicurezza tramite isolamento

• scaling orizzontale semplice
• bilanciamento del carico tra container
• limiti di risorse per container

• compatibile con pipeline CI/CD
• pronto per Kubernetes
• adatto al deployment cloud

---

Prerequisiti

• Docker installato (versione 20.10+)
• Docker Compose (opzionale, per setup multi-container)
• NVIDIA Docker (opzionale, per supporto GPU)
• conoscenze di base dei comandi Docker

Installare Docker

# Install Docker Desktop from docker.com
# Or using Homebrew
brew install --cask docker

sudo apt update
sudo apt install docker.io docker-compose
sudo systemctl start docker
sudo systemctl enable docker

Verificare l’installazione

docker --version
docker-compose --version

---

Dockerfile di base per Whisper

FROM python:3.10-slim

# Set working directory
WORKDIR /app

# Install system dependencies
RUN apt-get update && apt-get install -y \
    ffmpeg \
    git \
    && rm -rf /var/lib/apt/lists/*

# Install Python dependencies
RUN pip install --no-cache-dir \
    openai-whisper \
    torch \
    torchaudio

# Copy application code (if you have custom scripts)
# COPY . .

# Set default command
CMD ["whisper", "--help"]

Costruire l’immagine

docker build -t whisper:latest .

Eseguire il container di base

docker run --rm whisper:latest whisper --version

---

Dockerfile con server API

FROM python:3.10-slim

WORKDIR /app

# Install system dependencies
RUN apt-get update && apt-get install -y \
    ffmpeg \
    git \
    curl \
    && rm -rf /var/lib/apt/lists/*

# Install Python dependencies
RUN pip install --no-cache-dir \
    openai-whisper \
    torch \
    torchaudio \
    fastapi \
    uvicorn \
    python-multipart

# Create directories for audio and output
RUN mkdir -p /app/audio /app/output

# Copy application code
COPY app.py .
COPY requirements.txt .

# Expose API port
EXPOSE 8000

# Health check
HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
    CMD curl -f http://localhost:8000/health || exit 1

# Run API server
CMD ["uvicorn", "app.py:app", "--host", "0.0.0.0", "--port", "8000"]

Esempio di server API (app.py)

from fastapi import FastAPI, File, UploadFile
from fastapi.responses import JSONResponse
import whisper
import os

app = FastAPI()

# Load Whisper model (can be configured via env)
model_name = os.getenv("WHISPER_MODEL", "base")
model = whisper.load_model(model_name)

@app.get("/health")
def health():
    return {"status": "healthy"}

@app.post("/transcribe")
async def transcribe(file: UploadFile = File(...)):
    # Save uploaded file
    file_path = f"/app/audio/{file.filename}"
    with open(file_path, "wb") as f:
        content = await file.read()
        f.write(content)
    
    # Transcribe
    result = model.transcribe(file_path)
    
    # Clean up
    os.remove(file_path)
    
    return JSONResponse(content={
        "text": result["text"],
        "language": result["language"]
    })

requirements.txt

fastapi==0.104.1
uvicorn[standard]==0.24.0
python-multipart==0.0.6
openai-whisper
torch
torchaudio

---

Configurazione Docker Compose

docker-compose.yml

version: '3.8'

services:
  whisper-api:
    build: .
    container_name: whisper-api
    ports:
      - "8000:8000"
    volumes:
      - ./audio:/app/audio
      - ./output:/app/output
    environment:
      - WHISPER_MODEL=base
      - CUDA_VISIBLE_DEVICES=0
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s

  # Optional: Redis for queue management
  redis:
    image: redis:7-alpine
    container_name: whisper-redis
    ports:
      - "6379:6379"
    volumes:
      - redis-data:/data
    restart: unless-stopped

volumes:
  redis-data:

Eseguire con Docker Compose

# Start services
docker-compose up -d

# View logs
docker-compose logs -f whisper-api

# Stop services
docker-compose down

---

Supporto GPU con Docker

Installare NVIDIA Docker

# Add NVIDIA Docker repository
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | \
    sudo tee /etc/apt/sources.list.d/nvidia-docker.list

# Install
sudo apt-get update
sudo apt-get install -y nvidia-docker2
sudo systemctl restart docker

Dockerfile con supporto GPU

FROM nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04

WORKDIR /app

# Install Python
RUN apt-get update && apt-get install -y \
    python3.10 \
    python3-pip \
    ffmpeg \
    git \
    curl \
    && rm -rf /var/lib/apt/lists/*

# Install Python dependencies with CUDA support
RUN pip3 install --no-cache-dir \
    openai-whisper \
    torch \
    torchaudio \
    --index-url https://download.pytorch.org/whl/cu118

# Install API dependencies
RUN pip3 install --no-cache-dir \
    fastapi \
    uvicorn \
    python-multipart

COPY app.py .
EXPOSE 8000

CMD ["uvicorn", "app.py:app", "--host", "0.0.0.0", "--port", "8000"]

Eseguire con GPU

# Using docker run
docker run --gpus all -p 8000:8000 whisper-gpu:latest

# Using docker-compose

docker-compose.yml con GPU

version: '3.8'

services:
  whisper-api:
    build: .
    container_name: whisper-api-gpu
    runtime: nvidia
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
    ports:
      - "8000:8000"
    volumes:
      - ./audio:/app/audio
      - ./output:/app/output
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

---

Dockerfile ottimizzato per la produzione

# Multi-stage build for smaller image
FROM python:3.10-slim as builder

WORKDIR /app

# Install build dependencies
RUN apt-get update && apt-get install -y \
    gcc \
    g++ \
    git \
    && rm -rf /var/lib/apt/lists/*

# Install Python packages
COPY requirements.txt .
RUN pip install --user --no-cache-dir -r requirements.txt

# Final stage
FROM python:3.10-slim

WORKDIR /app

# Install runtime dependencies only
RUN apt-get update && apt-get install -y \
    ffmpeg \
    curl \
    && rm -rf /var/lib/apt/lists/* \
    && apt-get clean

# Copy Python packages from builder
COPY --from=builder /root/.local /root/.local

# Make sure scripts in .local are usable
ENV PATH=/root/.local/bin:$PATH

# Create non-root user for security
RUN useradd -m -u 1000 whisper && \
    mkdir -p /app/audio /app/output && \
    chown -R whisper:whisper /app

USER whisper

# Copy application code
COPY --chown=whisper:whisper app.py .
COPY --chown=whisper:whisper requirements.txt .

EXPOSE 8000

HEALTHCHECK --interval=30s --timeout=10s --start-period=40s --retries=3 \
    CMD curl -f http://localhost:8000/health || exit 1

CMD ["uvicorn", "app.py:app", "--host", "0.0.0.0", "--port", "8000"]

Vantaggi del build multi-stage

• Immagine più piccola – solo dipendenze runtime nell’immagine finale
• Build più veloci – cache separata per le dipendenze di build
• Sicurezza migliore – utente non root, superficie di attacco ridotta

---

Variabili d’ambiente

Dockerfile

FROM python:3.10-slim

WORKDIR /app

RUN apt-get update && apt-get install -y \
    ffmpeg \
    git \
    curl \
    && rm -rf /var/lib/apt/lists/*

RUN pip install --no-cache-dir \
    openai-whisper \
    torch \
    torchaudio \
    fastapi \
    uvicorn \
    python-multipart

COPY app.py .

# Environment variables with defaults
ENV WHISPER_MODEL=base
ENV MAX_FILE_SIZE=100MB
ENV LOG_LEVEL=INFO

EXPOSE 8000

CMD ["uvicorn", "app.py:app", "--host", "0.0.0.0", "--port", "8000"]

docker-compose.yml con variabili d’ambiente

version: '3.8'

services:
  whisper-api:
    build: .
    ports:
      - "8000:8000"
    volumes:
      - ./audio:/app/audio
      - ./output:/app/output
    environment:
      - WHISPER_MODEL=small
      - MAX_FILE_SIZE=200MB
      - LOG_LEVEL=DEBUG
      - CUDA_VISIBLE_DEVICES=0
    env_file:
      - .env
    restart: unless-stopped

File .env

WHISPER_MODEL=small
MAX_FILE_SIZE=200MB
LOG_LEVEL=INFO
CUDA_VISIBLE_DEVICES=0

---

Gestione dei volumi

docker-compose.yml con volumi

version: '3.8'

services:
  whisper-api:
    build: .
    ports:
      - "8000:8000"
    volumes:
      # Bind mount for development
      - ./audio:/app/audio
      - ./output:/app/output
      
      # Named volume for model cache (persists across containers)
      - whisper-models:/root/.cache/whisper
      
      # Config volume
      - ./config:/app/config:ro
    environment:
      - WHISPER_MODEL=base

volumes:
  whisper-models:
    driver: local

Vantaggi

• Cache modelli – download una volta, riuso tra container
• Persistenza dati – i file di output sopravvivono ai riavvii
• Configurazione – aggiornare le config senza ricostruire l’immagine

---

Health check e monitoraggio

Dockerfile con health check

FROM python:3.10-slim

WORKDIR /app

RUN apt-get update && apt-get install -y \
    ffmpeg \
    curl \
    && rm -rf /var/lib/apt/lists/*

RUN pip install --no-cache-dir \
    openai-whisper \
    fastapi \
    uvicorn

COPY app.py .

# Health check endpoint
HEALTHCHECK --interval=30s \
            --timeout=10s \
            --start-period=40s \
            --retries=3 \
            CMD curl -f http://localhost:8000/health || exit 1

EXPOSE 8000
CMD ["uvicorn", "app.py:app", "--host", "0.0.0.0", "--port", "8000"]

Endpoint health check

from fastapi import FastAPI
import whisper

app = FastAPI()
model = whisper.load_model("base")

@app.get("/health")
def health():
    try:
        # Quick test transcription
        return {"status": "healthy", "model": "base"}
    except Exception as e:
        return {"status": "unhealthy", "error": str(e)}, 503

---

Casi d’uso comuni

Caso 1: ambiente di sviluppo

version: '3.8'

services:
  whisper-dev:
    build:
      context: .
      dockerfile: Dockerfile.dev
    volumes:
      - .:/app
      - /app/__pycache__
    ports:
      - "8000:8000"
    environment:
      - WHISPER_MODEL=tiny
      - DEBUG=true
    command: uvicorn app.py:app --reload --host 0.0.0.0 --port 8000

Caso 2: produzione con coda

version: '3.8'

services:
  whisper-api:
    build: .
    ports:
      - "8000:8000"
    environment:
      - REDIS_URL=redis://redis:6379
      - WHISPER_MODEL=small
    depends_on:
      - redis
      - worker

  worker:
    build: .
    command: python worker.py
    environment:
      - REDIS_URL=redis://redis:6379
      - WHISPER_MODEL=small
    volumes:
      - ./audio:/app/audio
    depends_on:
      - redis

  redis:
    image: redis:7-alpine
    volumes:
      - redis-data:/data

volumes:
  redis-data:

Caso 3: setup multi-modello

version: '3.8'

services:
  whisper-fast:
    build: .
    ports:
      - "8001:8000"
    environment:
      - WHISPER_MODEL=tiny
      - PORT=8000

  whisper-balanced:
    build: .
    ports:
      - "8002:8000"
    environment:
      - WHISPER_MODEL=base
      - PORT=8000

  whisper-accurate:
    build: .
    ports:
      - "8003:8000"
    environment:
      - WHISPER_MODEL=large
      - PORT=8000

---

Best practice

1. Usare immagini base specifiche

FROM python:latest

FROM python:3.10-slim

2. Ridurre i layer

RUN apt-get update
RUN apt-get install -y ffmpeg
RUN apt-get install -y git

RUN apt-get update && apt-get install -y \
    ffmpeg \
    git \
    && rm -rf /var/lib/apt/lists/*

3. Usare .dockerignore

__pycache__
*.pyc
*.pyo
*.pyd
.Python
.env
.venv
venv/
.git
.gitignore
README.md
*.md
.DS_Store

4. Impostare limiti di risorse

services:
  whisper-api:
    build: .
    deploy:
      resources:
        limits:
          cpus: '2'
          memory: 4G
        reservations:
          cpus: '1'
          memory: 2G

5. Usare health check

HEALTHCHECK --interval=30s --timeout=10s CMD curl -f http://localhost:8000/health || exit 1

6. Utente non root

RUN useradd -m -u 1000 whisper
USER whisper

7. Cache dei modelli

volumes:
  - whisper-models:/root/.cache/whisper

---

Risoluzione dei problemi comuni

Problema 1: il container termina subito

# Check logs
docker logs <container-id>

# Run interactively to debug
docker run -it whisper:latest /bin/bash

Problema 2: GPU non disponibile

# Verify NVIDIA Docker
docker run --rm --gpus all nvidia/cuda:11.8.0-base-ubuntu22.04 nvidia-smi

# Check runtime
docker info | grep -i runtime

Problema 3: memoria insufficiente

# Increase memory limit
deploy:
  resources:
    limits:
      memory: 8G

Problema 4: download del modello lento

# Use volume for model cache
volumes:
  - whisper-models:/root/.cache/whisper

Problema 5: permesso negato

# Fix permissions in Dockerfile
RUN chown -R whisper:whisper /app

---

Ottimizzazione delle prestazioni

1. Precaricamento del modello

# Download model during build
RUN python -c "import whisper; whisper.load_model('base')"

2. Usare faster-whisper

RUN pip install --no-cache-dir faster-whisper

3. Multithreading

CMD ["uvicorn", "app.py:app", "--host", "0.0.0.0", "--port", "8000", "--workers", "4"]

4. Allocazione risorse

deploy:
  resources:
    limits:
      cpus: '4'
      memory: 8G

---

Considerazioni sulla sicurezza

1. Usare immagini base ufficiali

FROM python:3.10-slim  # Official Python image

2. Scansionare le vulnerabilità

docker scan whisper:latest

3. Aggiornare le immagini

FROM python:3.10-slim  # Use latest patch version
RUN pip install --upgrade pip

4. Limitare l’accesso di rete

services:
  whisper-api:
    build: .
    networks:
      - internal
    # No external ports if accessed via reverse proxy

---

Conclusione

1. Iniziare in modo semplice – con un Dockerfile di base
2. Usare Docker Compose – per stack multi-servizio
3. Abilitare la GPU – per le prestazioni in produzione
4. Seguire le best practice – sicurezza, ottimizzazione, monitoraggio
5. Testare a fondo – prima del deployment in produzione

---

Passi successivi

• Costruisci il primo container – inizia dal Dockerfile di base
• Aggiungi il supporto GPU – se hai GPU NVIDIA disponibili
• Configura Docker Compose – per l’intero stack applicativo
• Distribuisci sul cloud – servizi container (ECS, GKE, AKS)