Can the Whisper API file size limit be increased beyond 25 MB?

Below is a summary of the current situation and recommendations based on the documentation:

Current Limitations

• 25 MB File Limit: The Whisper API is designed to accept files under 25 MB. This is a hard limit that applies to all audio files uploaded for transcription.
• Splitting Files Carefully: The documentation recommends splitting files into chunks while taking care not to break audio mid-sentence, as doing so may lead to lost context in the transcription.

Options and Current Guidance

• No Direct Increase Option: At this time, there is no parameter or setting available to simply increase the 25 MB file size limit. This limit applies as a safeguard and ensures optimal performance of the transcription service.
• Audio Compression: You may consider using compressed audio formats (and ensure that they are one of the supported types: flac, mp3, mp4, mpeg, mpga, m4a, ogg, wav, or webm) to reduce the file size below the 25 MB threshold. This can help preserve the natural flow of sentences without the need for splitting.
• Preprocessing Strategies: While using external libraries like PyDub is a suggested method to split files dynamically into smaller chunks, the challenge you highlighted (maintaining sentence continuity) remains. A tailored preprocessing step that splits audio on natural pause points may help, though this requires additional logic.

Future Updates and Feedback

• No Announced Updates: As of now, there have been no public updates or announcements on increasing the file size limit directly. Any changes to such limits would likely come through platform-wide updates.
• User Feedback: Given that feedback from users is important for future improvements, you might consider reaching out to OpenAI support or participating in community discussions if this limitation significantly impacts your workflow. This user feedback can help prioritize adjustments or additional features in the future.

In summary, while the current 25 MB limit is fixed, exploring audio compression and smart preprocessing solutions are the recommended strategies until any potential future enhancements are announced.