Speech model
A model that turns speech into text, text into speech, or holds a spoken conversation directly.
Also: speech recognition, text to speech, voice model
Speech recognition converts audio to text; OpenAI's Whisper, released openly in 2022 and trained on 680,000 hours of audio, made accurate multilingual transcription widely available. Text-to-speech models generate natural voices. Newer voice assistants, such as GPT-4o's voice mode and Gemini Live, process audio directly rather than passing through text, so they can respond quickly and hear tone. In India, AI4Bharat at IIT Madras and Sarvam AI have built speech models for Indian languages, and the government's Bhashini platform offers speech and translation services.