Rev has launched its benchmark report, The 2024 State of ASR, which underscores that accuracy is the cornerstone of automatic speech recognition (ASR) technology. 

Accurate transcription is essential for capturing critical moments, from breaking news to legal proceedings, and provides credibility to the situation at stake. However, the quality of an ASR-powered transcript can vary depending on audio conditions, speaker properties, and the ASR provider in use. The State of ASR report emphasizes the need for reliable and accurate ASR solutions, particularly in high-stakes scenarios.

For over 12 years, Rev has been collecting and transcribing data to train ASR models. This robust data set ensures that Rev has created an ASR that delivers precise transcriptions, no matter how challenging the audio environment. Our commitment to researching and implementing better ASR technology reflects our commitment to the industries most in need of accuracy—even when they don’t have perfect audio. 

How ASR Accuracy Was Tested Across Leading Providers

We partnered with an independent benchmarking service to deliver objective evaluations of not just our ASR but eight other leading ASR providers: AssemblyAI, AWS, Deepgram, Google, Microsoft, OpenAI, Otter, and Speechmatics. The study was designed to capture a wide range of recording scenarios and media formats, reflecting the diverse applications of AI-powered speech technology.

With more than 13 hours of test data, ASR providers were assessed for their word error rate (WER) across various audio domains and environments, such as noise robustness, far-field settings, and telephony. WER is, most simply put, the ratio of the number of errors in a transcript to the total number of words spoken. It’s the research method most commonly used to determine how well an ASR system comprehends spoken words and translates them into text.

Who Leads in Accuracy And Reliability

Rev performed better than any other ASR provider across the three tested audio domains: news and political speech, music and entertainment recordings, and legal proceedings. Rev’s model also demonstrated the most robust performance in noisy environments. It produced transcripts with nearly 1.5 fewer errors per segment than the next best competitor.

Overall, Rev’s low WER confirms its frontrunner status in the field, surpassing Amazon (AWS) by 5.6%, Otter by 24.1%, and Google by an impressive 60.5% in ASR accuracy. 

Driving Innovation and Improvement With ASR

These are just a few of the key findings from Rev’s The 2024 State of ASR Report. Read the study for a more comprehensive overview of the strengths and limitations of the industry’s top ASR technologies. The future of communication belongs to accurate and reliable ASR solutions. Rev is committed to championing accuracy at the forefront of this revolution. 

Download Now