NASA needed highly accurate speech-to-text to transcribe noisy space-to-ground communications, where traditional ASR systems failed to reach the required 80% word recognition rate. After testing multiple providers and even building with Kaldi, NASA selected Deepgram for its tailored speech models, achieving 86–87% accuracy. Deepgram now supports four use cases: ISS communications, Neutral Buoyancy Lab training, the IRIS medical chatbot running on Raspberry Pi, and robotic process automation. The result is searchable mission audio, reduced manual transcription, improved error detection, and more efficient mission workflows.

Join for free to read