Resources required
Running Storyteller Locally
AI-based audio transcription is a fairly resource-intensive task, so it will not run well on all hardware. Before going further, take a moment to reveiw the minimum resource required to run Storyteller to make sure that you have a machine that will be able to run Storyteller!
If you know you have what it takes, skip ahead to instructions to get up and running!
Minimum resources required
Here is a table describing some requirements for running the Storyteller backend:
| Component | Requirements | Notes |
|---|---|---|
| CPU | Intel/AMD or ARM64 — Up to 4 cores | Storyteller will use up to 4 cores of CPU. Storyteller does run on Apple Silicon! It will also technically run on other ARM devices, like a Raspberry Pi, but will likely be too slow to be usable. |
| GPU | NVIDIA (CUDA 11.8, 12.x, 13.x), newer AMD GPUs (RDNA 2, RDNA 3), Intel Arc GPUs, or Intel iGPUs | A GPU is not required to run Storyteller. See the docs on GPU acceleration |
| Storage | 10 GB (30 GB for CUDA) | This is the bare minimum necessary to run. Though final aligned EPUBs are ~300MB, you should budget ~1GB per book to accomodate source files and assets. |
| Memory | 8 GB | If using Docker Desktop for Windows or Mac, make sure that your Docker VM (or WSL2 VM, if using WSL2) is configured with at least 8GB of memory |
| Swap | ~12 GB (?) | If you find that transcription processes are dying unexpectedly, and you can't increase the amount of memory available, increasing swap can keep processes from being killed, though it may slow down processing |
Tips & tricks for low-resource machines
If your Storyteller server spontaneously crashes during alignment, the most likely culprit is that the process ran out of available memory. Here are some tips to try to work around this issue:
- If you’re running Docker Desktop on Windows or macOS, ensure that your Docker VM has access to as much memory as you can give it. The default on Windows is a measly 2GB! You can adjust this in the Docker Desktop settings.
- If you’ve already given the Docker VM access to as much memory as you can give it, you can try to increase the amount of swap. Swap is disk storage that processes can use to save pages of memory when there isn't enough available. This is much slower than just having more memory, but it means that you process is less likely to outright crash!
- If you’re maxed out on memory and swap, try reducing the maximum track length in the audio settings.
Other Hosting Options
PikaPods and ElfHosted are services for running open source apps like Storyteller. They are paid services, and they share profits with open source maintainers, so every PikaPods and ElfHosted instance helps support future Storyteller development.
PikaPods
PikaPods does not have a landing page for Storyteller. You can either search for it, or it is pretty easily found on their media page.
Elfhosted
Elfhosted's Storyteller page