Run Whisper transcription behind your own API
Deploy Faster-Whisper as an autoscaling endpoint that takes an audio URL or base64 file and returns text. Beam manages the GPUs and bills per second of actual transcription.
Scalable audio infrastructure
Run Whisper with warm models, async queues for long audio, and webhooks when jobs land.
Models cached on volumes
Download large-v3 once to a persistent volume, and every container after that loads from disk, not the Hub.
Explore docsScale with your backlog
Endpoints scale out as transcription requests pile up and back to zero when the queue drains.
Explore docsAsync for long audio
Hour-long recordings don't fit a synchronous request. Managed task queues take the job and return a task ID instantly.
Explore docsWebhooks when done
Attach a callback_url and Beam POSTs to your app the moment a transcript is ready.
Explore docsRight-size the GPU
An A10G comfortably serves Whisper large-v3, and you can drop to a T4 for smaller models by changing one line.
Explore docsPer-second billing
Pay for seconds of transcription compute and nothing while the endpoint sits idle.
Explore docsTranscribe in three steps
Deploy the endpoint
Wrap Faster-Whisper in an @endpoint with on_start, so the model loads once per container and stays warm.
Send audio
POST a URL to an audio file or a base64-encoded payload. The endpoint handles either.
Get text back
Synchronous requests return the transcript directly; long jobs return a task ID you can poll or receive via webhook.
$ beam deploy app.py:transcribe
=> Deployed 🎉
=> https://faster-whisper-abc123.app.beam.cloud
$ curl -X POST https://faster-whisper-abc123.app.beam.cloud \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer YOUR_TOKEN' \
-d '{"url": "https://example.com/meeting.mp3"}'
{"text": "Welcome back to the show..."}Frequently asked questions
What audio inputs are supported?
Pass a URL to an audio file or send the file itself base64-encoded. The Whisper example handles both patterns.
How do I handle hour-long recordings?
Deploy the same function as a task queue instead of an endpoint. Requests return a task ID immediately, and a callback URL notifies your app when the transcript is ready.
Which languages does it support?
Whisper is multilingual. Pass a language parameter to pin the language, or let the model auto-detect.
Can I use a different speech model?
Yes. Anything you can run in Python works the same way. You can swap Faster-Whisper for WhisperX, a distilled variant, or your own fine-tuned speech model.
What does an idle transcription API cost?
Nothing. Containers scale to zero between jobs and billing is per-second while transcription runs.
Related use cases
Run text-to-speech behind a managed API
Serve open TTS models with hosted audio outputs.
Task QueuesFan out batch jobs across thousands of containers
Fan out batch inference and ETL across thousands of containers.
SandboxesGive your AI agents a safe place to run code
Isolated sandboxes for agent-generated code, booting in seconds.
Give your agents their own machines
Sign up, install the CLI, and launch a sandbox, VM, or GPU from your terminal or your coding agent.