Track: VoiceTech |
| The path to end-to-end spoken language redaction |
| While redaction of sensitive information can be a relatively simple natural language understanding (NLU) problem when applied to written word, the application of these same methods to speech is more complicated. Typical approaches rely on automatic speech recognition (ASR) to reduce the problem to text, which may fail to completely redact information due to ASR errors. We propose an end-to-end model, applied directly on speech, in order to create an appropriate mask on sensitive information. |
|
| Presentation Video |
