Content Limits for Audio Processing
What is this page about?
To create transcripts, the Secure Academic Interview & Speech Transcriber application uses Google's advanced artificial intelligence service (Gemini). This service has built-in, non-overridable safety filters that always refuse to process certain types of content — regardless of the settings we ourselves apply.
These limits come from the AI provider's side, and we cannot disable or override them. Below we explain clearly which content falls into this category, what happens in such cases, and how we handle your recording.
Important perspective: The vast majority of uploaded recordings are not affected by any restriction. The cases described here are rare edge cases, not normal operation.
What content does the system always refuse?
The following categories are subject to so-called non-configurable filters. The AI provider blocks these automatically, entirely independently of our settings.
1. Child sexual abuse material
The strictest and unconditional prohibition. Processing of any recording containing such material is refused under all circumstances, without exception. No setting can disable this filter.
2. Sensitive personally identifiable information
If the audio contains certain sensitive data spoken aloud — for example a full payment card number, banking login credentials, a password, or a government identifier (national ID, social security / tax number, passport number) — the system may interrupt processing. This filter is designed to protect the individuals appearing in the recording.
3. Verbatim reproduction of copyrighted text
If the recording contains a lengthy, word-for-word reading of a copyrighted work, processing may be halted. For speech and interview transcripts this rarely occurs.
4. Blocklisted terms
Terms that appear on the AI provider's internal, non-public blocklist may likewise interrupt processing.
What is NOT a reason for refusal?
It is important to be clear: the system is built to transcribe real, everyday human speech. Heated arguments, profanity, strong emotions, sensitive or adult topics, and political or ideological differences do not, in themselves, cause a refusal.
The tool does not judge the tone, intensity, or subject matter of a conversation — it filters only the narrowly defined categories listed above.
What happens if the system refuses a recording?
- The transcript is not produced, and you receive a clear error message.
- The credits charged for the operation are automatically refunded to your balance. We do not charge for a refused job.
- After the processing attempt, the uploaded audio file is automatically and permanently deleted from our server.
How do we handle your recording? (Privacy)
- Content filtering takes place on the AI provider's side, in an automated manner. No human reviewer listens to or examines the recording during this process.
- We do not store the audio file beyond processing: it is deleted whether the job succeeds or fails.
- Data processing takes place within the European Union.
Your responsibility
By using the service, you agree that:
- you will only upload audio that you have the right to process, or for which the individuals in the recording have given consent;
- you will not upload unlawful content, or confidential or intimate recordings made without consent;
- you will comply with Google's Generative AI Prohibited Use Policy, which underlies the service: policies.google.com/terms/generative-ai/use-policy
False positives
Because this is an automated system, it may rarely flag a perfectly harmless recording in error. This is not a deliberate rejection or content censorship. If this happens, we recommend trying the upload again.
This notice provides general information about the technical limits of audio processing and does not constitute legal advice. These restrictions may change in accordance with the AI provider's policies in effect at any given time.