Audio Transcriber

A basic project to make use of the OpenAI Whisper general-purpose speech recognition model to transcribe audio files.

Reference: https://github.com/openai/whisper

The transcriber function expects to be passed a model name to use, a blob container of input audio files, and (optionally) an output blob container for transcripts.

Installation

Dependencies for this project are managed using uv. With uv installed, change into the project directory and run:

uv sync

Activate the virtualenv like so:

source .venv/bin/activate

To run Python commands in the activated virtualenv, thereafter run them like so:

python manage.py

Manage new or updated project dependencies with uv also, like so:

uv add newpackage==1.0

Environment variables

This project uses python-dotenv to set environment variables (in a .env file). The following variables are required for the project to run:

AZURE_CONNECTION_STRING=AzureStorageAccountConnectionString

Running

Run locally like so:

python transcriber.py --help

Generate a transcript output for a local file like so:

import os
from pathlib import Path
from transcriber import get_model, get_transcription, write_transcription

# Assuming audio files are saved to 'data':
data_path = Path(os.curdir, "data")
input_file_path = os.path.join(data_path, '17-7-25 kps13 p151 (obs 1).m4a')

# Instantiate the model, generate and write the transcription (TSV, transcript directory).
model = get_model("small.en")
transcription = model.transcribe(audio=input_file_path, initial_prompt="Field observations of pollinator insects:", language="en")
write_transcription(transcription, "17-7-25 kps13 p151 (obs 1)")

Docker image build

Build and push a multi-architecture-capable image from the Dockerfile like so:

docker buildx create --name container-builder --driver docker-container --use --bootstrap
docker buildx build --push --platform linux/amd64,linux/arm64 -t ghcr.io/dbca-wa/audio-transcriber .