Today we're releasing our biggest accuracy improvement yet for phone call transcription, making AssemblyAI the most accurate Speech to Text API for transcribing phone calls.
Here are some examples showing how we compare to Google's Speech-to-Text API, often touted as one of the more accurate APIs for automated transcription on the market. We did not train our models on any of the below samples -- they come from a holdout test dataset used for benchmarking.
Out of the box, you can now use AssemblyAI to get extremely accurate transcriptions for your phone call recordings. And to get even higher accuracy, you can take advantage of our customization features to boost accuracy for keywords and phrases, and to add custom vocabulary, for unparalled accuracy.
Our team's research into more advanced deep learning technology like better data augmentation techniques, more robust neural network architectures, and higher accuracy on compressed audio, have gotten us to this milestone. That being said, our team continues to push forward because we have no plans to stop improving. This is why, when you choose AssemblyAI, you can have confidence that you will always have the most advanced and accurate Speech-to-Text available on the market powering your product.
We'll be posting more thorough benchmarks, comparing our API to other APIs like Google and AWS Transcribe, in the coming weeks.
To get started with AssemblyAI, you can get an API token here. And if you need any help, our team of developers are available over email or Slack whenever you need us.