ASR systems treat speech variations as speech anomalies. By training engines on authentic speech patterns from Deaf and Hard of Hearing speakers, GoVoBo is closing the performance gap.
Most voice models are trained on default vocal pitches, ignoring the unique phonetic variables and cadence of DHH voices. This leads to high transcription error rates, which lock DHH users out of automated speech tools.
Working alongside Gallaudet University and AppTek, GoVoBo Labs is actively building a representative voice database. The study gathers anonymized spoken sentences to train acoustic models to listen accurately, ensuring speech intelligence works for everyone.