- Most accurate transcription for our use case. - Very easy API. - Good customisability on the formatting of results. - Competitve pricing.

low latency and coverage for most languages

Comparison with Amazon Transcribe: We started with Amazon (AWS) Transcribe thinking it must be the best. After spending a lot of money (it is many fold more expnsive than Deepgram), patience (its inferior in quality) and time, we gave up. Running whisper on an AWS machine: We decided to run Whisper on an Amazon machine. Afterall Whisper is free and we would only pay for compute. The quality was definitely better. But we soon realized that with T@ large ($500/month) we were experiencing very high latency during peak hours of the day and we were still paying for the idle compute at remaining hours. Switch to Deepgram: After switching to Deepgram we found out that not only they are high quality, they are faster (latency of less than a minute for audio files upto an hour) , cheaper (no need for idle compute, only pay per use), and same quality as whisper with one of their models we chose called Nova (they also offer two other models including Whisper and one more). Now we are exploring their real time transcription, LLM for real time analysis of streams, and also infrastructure needed to develop voice bots. I highly recommend Deepgram to anyone!