[investigate/prototype] speech_to_text_generation_service approach 2: Explore AWS SageMaker

[jmartin-sul]

Keep in mind the guidelines at the top of #1. In particular, the ones about ultimately using Terraform for deployment, and about access to model configuration.

Note: some naming might change, depending on terminology decisions, see https://github.com/orgs/sul-dlss/projects/65/views/1?pane=issue&itemId=79627337

SageMaker is "a fully managed machine learning service". Whisper is a model that can be used with it. see:

• https://docs.aws.amazon.com/sagemaker/latest/dg/whatis.html
• https://aws.amazon.com/blogs/machine-learning/host-the-whisper-model-on-amazon-sagemaker-exploring-inference-options/

Questions to determine whether this is a viable option

• Is it available via Cardinal Cloud? Is there a clear setting or API parameter or similar that we trust will prevent SageMaker from using our data as training data?
• Does it expose the Whisper configuration parameters we care about?
• What's it cost relative to e.g. a Docker container that we run in ECS? Is it acceptably affordable?

[jmartin-sul]

blocked till we can get permissions issues straightened out

[jmartin-sul]

maybe more for building and tuning models, and less for deploying, but we'd like to confirm this.

was in blocked due to AWS perms issues, but we think this is actionable now.