Custom Models

    What is a Model

    A "model" is an algorithm that can convert data into a different representation. In our case, that means converting audio into text. When we use the word "model" we are talking about the algorithm we train on millions of audio files which can convert audio into text.

    Acoustic Models

    An Acoustic Model learns about audio characteristics like accent, background noise, recording quality, recording environment, etc. We have a number of Acoustic Models available, and are always launching more, that are better at different types of data.

    Model Use Case Availability
    Default Best for all types of data and accents. English only. Public
    Australian-English Tailored for Australian-accented English. You can expect a 2-3% improvement compared to the default model on Australian data. Public
    UK-English Tailored for UK-accented English. You can expect a 2-3% improvement compared to the default model on UK data. Coming Soon
    Indian-English Tailored for Indian-accented English. You can expect a 2-3% improvement compared to the default model on Indian-English data. Coming Soon
    South East Asian-English Tailored for South East Asian-accented English. You can expect a 2-3% improvement compared to the default model on South East Asian-English data. Coming Soon
    Custom Tailored for your data. You can expect a large improvement compared to the default model on your data. Contact us for more info: support@assemblyai.com

    Language Models

    The Language Model is used to help disambiguate between probable words and phrases. For example, did you say "two" or "to" or "too"?

    You can use the API to train a custom Language Model which will include any custom vocabulary, like product names and person names, unique to your application. You can also boost accuracy for common phrases and speaking patterns. For example, speaking patterns in a conversation versus a lecture/presentation are very different, and a Language Model can capture this difference.