Whisper-STT CLI in Docker

README.md

Usage

1. Download Docker for Desktop
2. Create an audio file (.wav) and save it in your Desktop folder. Example: ~/Desktop/sample-audio.wav
3. Build the Docker image and transcribe the audio

docker build -t whisper-app .
docker run -it --rm -v ~/Desktop:/app/audio whisper-app python transcribe.py /app/audio/sample-audio.wav

4. You may load a different model in the transcribe.py and rebuild the docker image and re-run the docker container.

Dockerfile

FROM python:3.9-slim

WORKDIR /app

COPY . /app

RUN pip install --no-cache-dir -r requirements.txt \
    && apt-get update \
    && apt-get install -y ffmpeg \
    && useradd -m appuser \
    && chown -R appuser /app

USER appuser

EXPOSE 8080

CMD ["python", "transcribe.py"]

requirements.txt

numpy
soundfile
pydub
openai-whisper
setuptools-rust

transcribe.py

import numpy as np
import io
import soundfile as sf
import whisper
import os
from pydub import AudioSegment


print("loading model ...")
model = whisper.load_model("tiny")
print("model loaded.")

def transcribe(audio_file):
    if not os.path.exists(audio_file):
        print("No audio provided")
        os.exit(1)
      
    # Read the audio file with pydub
    audio = AudioSegment.from_file(audio_file)
    audio = audio.set_channels(1)
    audio = audio.set_frame_rate(16000)
    
    # Convert audio to WAV format
    wav_io = io.BytesIO()
    audio.export(wav_io, format="wav")
    wav_io.seek(0)
    
    # Use soundfile to read the WAV data
    audio_data, sample_rate = sf.read(wav_io)
    audio_array = np.array(audio_data, dtype=np.float32)
    
    # Transcribe the audio with Whisper model
    result = model.transcribe(audio_array)
    transcript_text = result["text"]
    
    return transcript_text


# Read the audio file path name frm the command line
print("Usage: python whisper.py <audio_file>")
audio_file = os.sys.argv[1]
print(transcribe(audio_file))