Traditional cloud-based speech recognition services (such as Google Cloud Speech-to-Text and AWS Transcribe) require audio to be uploaded to remote servers for processing, posing privacy risks including data transmission security, cloud-side leakage, and use by service providers for model training.