Role of LLMs in WER reduction

At Ukumi, Transcription is one of the foundation building blocks. Getting it right, getting is fast and getting it in many languages is important.

Sachin from Ukumi, shared some of the open problems in transcription space at Norwegian AI Research Consortium event on transcription. Link here.

At Ukumi, we benchmarked many transcription services and we end up loving Deepgram. In our investigtations we have found the service reliable, SOTA in WER and zippingly fast.

We would love to know from you all, where you are using transcription in your workflows, what do you think are your usecases which combine LLMs and Transcription.

We find it interesting in reducing word error rate further with context shared with human in loop workflows.

We love to hear from you and learn about your journey and learnings linked with ASR models and transcription.