Deepgram reports that its speech models are used in production by over 5,000 organizations and 200,000 developers, having processed over 50,000 years of audio.