Keyterm Filtering API

Keyterm Filtering APIReduce false positives in speech-to-text transcription by phonetically verifying keyterms before passing them to your STT service.The Problem: Keyterm OverexpressionWhen you provide keyterms to STT services (like Deepgram) to boost recognition of specific words, the service may incorrectly insert those keyterms when phonetically similar words are actually spoken. For example: audio says "policy" but STT outputs "police" because "police" was provided as a keyterm.The Solution: Pre-STT Phonetic FilteringAditu's keyterm filter sits before your STT service. It uses phonetic analysis (wav2vec2 CTC) to verify which keyterms are actually present in the audio, then only passes verified keyterms to your STT provider. Server-side processing takes approximately 15ms.API EndpointsPOST /keyterms/registerRegister keyterms for a session. Provide a session ID, language code, and list of keyterms. Supported languages: en-IN, hi-IN.POST /keyterms/filterSend audio with a session ID. Returns which registered keyterms were detected in the audio, with similarity scores and timing breakdown.WebSocket /keyterms/ws/{session_id}Stream audio over WebSocket for real-time keyterm filtering. Server-side VAD detects utterance boundaries and returns filter results automatically. Compatible with Deepgram Flux mid-stream Configure for live keyterm updates.GET /keyterms/session/{session_id}Retrieve registered keyterms for a session.DELETE /keyterms/session/{session_id}Delete keyterms from a session.Base URLhttps://gpu.aditu.techPricingCurrently free while in early access. No API key required.