AWS Bedrock fails to ingest .SRT files

Question

My team is making a product where we want users to upload video transcript files to a Bedrock knowledge base. The file format we're working on right now .SRT, which is just a plain text file under the hood. Here is some documentation on the format of .SRT files: https://www.rev.com/resources/what-is-an-srt-file-format-create-use-srt-files

I've tried converting the .SRT file to a .TXT file and setting the MIME type to "text/plain", but Bedrock isn't able to ingest it, citing the following reason: "Encountered error: Ignored 1 files as their file format was not supported." This happens even when I touch a .TXT file and enter the .SRT content in it.

Through some trial and error I've found the problem is in the content of the file--something about what is normally in a .SRT file makes it unable to be ingested. Does it have something to do with the amount of "natural language" in the file compared to the amount of metadata (like timestamps)?

Matthew D · Answer

Did your problem get resolved?

In addition the following may help.

* Trying different file formats that Bedrock explicitly supports (beyond just converting to .txt)
* Validating the text encoding of the file (ensuring it's UTF-8 or another supported encoding)
* Checking if there are any special characters or formatting in the SRT file that might be causing issues

re:Post Agent · Answer

When AWS Bedrock Knowledge Base fails to ingest files with the error "Ignored files as their file format was not supported," the issue may not be with the file extension itself but rather with the content structure or formatting within the file.

SRT files contain a specific format with numbered entries, timestamps, and transcript text. Even when converting to .TXT format, this structure remains and might be causing the ingestion issue. Bedrock Knowledge Base has specific requirements for supported file formats and content.

There are several potential solutions to try:

1. Preprocessing the file: Remove the numbered entries and timestamps, keeping only the actual transcript text. This would transform the content into more natural language text that Bedrock might better recognize.

2. Enable chunking: If you're using "No chunking" as your strategy, try enabling chunking instead. This can help process files by breaking them into smaller, more manageable sections.

3. Check file size: Ensure your file isn't exceeding size limitations. Even files with only 600 words have been reported to cause errors when chunking is disabled.

4. Verify permissions: Ensure that the Bedrock service has the necessary permissions to access your S3 bucket where the files are stored.

5. Split larger files: If your transcript files are lengthy, consider breaking them into smaller files to stay within processing limits.

The issue is likely related to how Bedrock processes the structured content within SRT files rather than the file extension itself. The timestamps and entry numbers may be interfering with Bedrock's ability to process the content as natural language text.  
**Sources**  
[Issue related to AWS Bedrock Knowledge Base - File format not supported | AWS re:Post](https://repost.aws/questions/QUzXcDbelsR0qZg-jbr-oY_Q/issue-related-to-aws-bedrock-knowledge-base-file-format-not-supported)  
[Resolve file sync failures with Amazon Bedrock Knowledge Bases | AWS re:Post](https://repost.aws/knowledge-center/bedrock-resolve-file-sync-failures)  
[Bedrock KB data source sync error with no chunking and file of 600 words | AWS re:Post](https://repost.aws/questions/QUYut3fwJ9SCe7kfNP76pdMg/bedrock-kb-data-source-sync-error-with-no-chunking-and-file-of-600-words)  
[Bedrock Knowledge Base Creation: Failure to Ingest from S3 | AWS re:Post](https://repost.aws/questions/QU6igZfa1TQGiq0geGB1NIxQ/bedrock-knowledge-base-creation-failure-to-ingest-from-s3)

AWS Bedrock fails to ingest .SRT files

Add your answer

Relevant content