gcp-transcribe

Cloud-based MP3 transcription CLI built on Google Speech-to-Text v2 (Chirp 2). Submit a job, get a job ID back, then poll for status and fetch results when ready.

• Async batch workflow for long-form audio
• Multilingual via Chirp 2 (any BCP-47 language Chirp 2 supports)
• Outputs .txt, .json, and optional .srt
• Jobs survive across CLI invocations and reboots

Prerequisites

• Python 3.10+
• A Google Cloud account with billing enabled
• The gcloud CLI installed locally

Set up Google Cloud

You can do this in the Google Cloud Console or entirely from the terminal. The CLI flow:

1. Create or select a project

# Create a new project
gcloud projects create <your-project-id> --name="Transcribe"

# Or pick an existing one
gcloud projects list
gcloud config set project <your-project-id>

2. Enable billing

A billing account must be linked to the project. List your billing accounts and link one:

gcloud billing accounts list
gcloud billing projects link <your-project-id> \
  --billing-account=<billing-account-id>

(Or do this once in the Console under Billing → Link a billing account.)

3. Enable the required APIs

gcloud services enable \
  speech.googleapis.com \
  storage.googleapis.com \
  --project=<your-project-id>

4. Authenticate

This tool uses Application Default Credentials, so you don't need a service-account key for local use:

gcloud auth application-default login
gcloud auth application-default set-quota-project <your-project-id>

The user account you log in with needs roles that allow Speech-to-Text recognition plus bucket and object management on the staging bucket. For personal use, Owner or Editor on the project is sufficient. For tighter scoping, the minimum is:

• roles/speech.client
• roles/storage.admin (or storage.objectAdmin + storage.bucketAdmin if you'd rather not grant full admin)

Running in CI or a server? Create a service account, grant the same roles, download a JSON key, and point GOOGLE_APPLICATION_CREDENTIALS at it instead of running gcloud auth ....

Install

pip install gcp-transcribe

Or from source:

git clone https://github.com/bengeos/gcp-transcribe.git
cd gcp-transcribe
pip install -e .

Configure

Create a .env file in your working directory (or export these as environment variables):

GCP_PROJECT_ID=your-gcp-project-id
GCP_REGION=us-central1
GCS_BUCKET=your-staging-bucket
DEFAULT_LANGUAGES=en-US

Verify everything is wired up:

transcribe doctor

Usage

Submit a job

transcribe submit audio.mp3
# → prints a job ID

Multiple files in one job:

transcribe submit a.mp3 b.mp3 c.mp3

Override languages per call:

transcribe submit audio.mp3 -l en-US -l es-ES

Check status

transcribe status <job-id>

State is one of running, done, failed, fetched.

Fetch the transcript

transcribe result <job-id> --out ./transcripts

Adds subtitles:

transcribe result <job-id> --out ./transcripts --srt

GCS staging objects are deleted after a successful fetch. Use --keep-remote to retain them.

Other commands

transcribe list                # all known jobs
transcribe forget <job-id>     # remove a job from the local manifest
transcribe doctor              # verify config + GCP auth

Costs

Chirp 2 batch transcription is roughly $0.016 per minute of audio. A 3-hour file ≈ $2.88.

Limitations

• No speaker diarization (not supported by Chirp 2 in batch mode).
• Each file is capped at the Speech-to-Text v2 batch limit (currently 8 hours per file).

License

MIT