AssemblyAI Research
and Updates

A Survey on End-To-End Speech Recognition Architectures in 2021

As we continue our core research and development efforts to push the boundaries of state of the art accuracy, we begin exploring architectures for end-to-end speech recognition that are gaining relative new popularity, not only in the research domain but also in production settings in industry. We are specially interested in architectures that have shown production level accuracy matching or surpassing that of conventional hybrid DNN-HMM speech recognition systems. With that in mind, we survey the following two architectures: Listen Attend and Spell, and Recurrent Neural Network Transducers.

Building an end-to-end Speech Recognition model in PyTorch

In this blog post, we'll walk through how one would build their own end-to-end speech recognition model in PyTorch. The model we'll build is inspired by Deep Speech 2 (Baidu's second revision of their now-famous model) with some personal improvements to the architecture. The output of the model will be a probability matrix of characters, and we'll use that probability matrix to decode the most likely characters spoken from the audio.

PII Redaction and Accuracy Improvements

Last month, we introduced over a dozen new PII Redaction Policies as part of a big overhaul to our PII Redaction feature to make it more flexible and customizable. Now you can specify exactly what you want redacted from your transcriptions.

WhatConverts | AssemblyAI Case Study

Since switching to AssemblyAI, WhatConverts has noticed a significant improvement in their call transcription accuracy, making it easier for customers to identify their best leads. 

AssemblyAI Wins Best Public API 2020 🏆

This week Nordic APIs announced AssemblyAI Speech-to-Text as the Best Public API in 2020! Nordic APIs holds this competition each year as one of the most trusted sources in API reviews and recommendations.

PII Redaction Policies for Speech-to-Text

To best-fit AssemblyAI's PII Redaction to your application, you can now select from a set of redaction policies when PII Redaction is enabled. You can include any or some of these policy names when submitting audio or video files for transcription.

Speaker Diarization: Speaker Labels for Mono Channel Files

Speaker diarization (diarisation) is the process of splitting audio or video inputs automatically based on the speaker's identity. It helps you answer the question "who spoke when?".

Voice Search Partnership with Algolia

We’re happy to announce that we’ve teamed up with Algolia to launch a plugin for their popular InstantSearch.js library. With this library, developers can add accurate Voice Search to their websites in just a few lines of code.

[Webinar] Conversation Intelligence with CallRail

Join CallRail’s Senior Product Manager, Christina Bourne, and Demand Gen Marketing Manager, Rachel Ward, along with AssemblyAI’s CEO, Dylan Fox. Together, they’ll talk through 5 ways sales and marketing professionals are using AI to clear the path to purchase.

Teaming up with Algolia for effortless Voice Search

Did you know that 71% of users prefer to search through voice compared to using a keyboard? If you’ve started dictating more with your mobile device and/or laptop, especially to do search, you’ll understand why - talking is much faster than typing!

Improved WER - Speech-to-Text Accuracy vs. Google, AWS (May 2020 Update)

With our most recent model update, we are continuing to out-perform other API's like Google Cloud's Video Model, AWS Transcribe, and Microsoft Azure in accuracy. Read more to see the full report, and to read about other product updates launched this month!

Automated SRT and VTT Video Captions (April 2020 Update)

Now you can easily export your transcription in SRT or VTT format, to be plugged into a video player for subtitles and closed captions!

PII Redaction for Speech-to-Text Transcriptions (March 2020 Update)

Our PII Redaction feature automatically detects and removes sensitive information, like credit card and social security numbers, from the transcription text. These sensitive numbers are replaced with pound signs "#" in the transcript.