The interactive demo below demonstrates how nCompass’s audio denoising tool can enable transcription of audio with multiple overlapping voices with existing Speech-to-Text (STT) tools like Deepgram. Current STT tools such as Deepgram do not remove overlapping background voices and thus produce transcriptions that are unintelligible.

When you upload any audio file of your choice (‘.wav’ format) we do 3 things:

  1. Use Deepgram’s API to transcribe your audio clip
  2. Run your audio clip through our audio denoiser
  3. Transcribe the denoised audio using Deepgram’s API

If you run either of our two example files, you can see that Deepgram’s transcription picks up a lot of the words in the background as well. This makes the transcript innacurate and incomprehensible. On the other hand, with the denoised output, the transcription is exactly what the main speaker is saying!