Voice transcription tools promise the quick and easy conversion of speech into text, but this efficiency can come at the cost of accuracy. Matthew Rodano explains when speech-to-text technologies might help and when they might hinder your transcription projects.
As voice recognition technology has improved, so has the use of speech-to-text programs to transcribe audio files in personal, academic, and business settings. Depending on the subject matter or circumstances, many voice transcription tools can now deliver on their promises of quick and easy conversion of speech into text. But what about their promises of accuracy?
Different speech-to-text programs have different levels of ability and complexity, but even those that claim to use artificial intelligence (AI) or machine learning (ML) can struggle to produce quality output for use in many professional business settings. They might provide a rough draft or workable first pass for basic transcription tasks, thereby saving the writer or reviewer transcription time. However, such tools are unlikely to be appropriate for use in highly technical industries, such as the legal or medical fields, which use more complex and challenging terminology.
Depending on the language and how the transcription is to be used, the same can also apply to industries that might be considered less “technical,” such as media and entertainment. As the high-profile coverage of the subtitles in the 2021 breakout Korean-language drama Squid Game recently illustrated, some viewers will not compromise their expectations when it comes to the quality of subtitling (or voice-over) content.
Therein lies another problem for many speech-to-text transcription tools. While they might work to transcribe audio files in English and other Western languages adequately, current speech-to-text technology is known to struggle in languages such as Mandarin, Japanese, Korean, Arabic, and many others.
When and How to Use Audio Transcription Technology
Whether or not you can rely on speech-to-text software depends on the technicality and complexity of your industry and its terminology, therefore—as well as on the language(s) you are looking to transcribe. Of course, it is also reliant on the quality of the original audio file and the clarity of the speaker.
Even where the technology can provide a good approximation from spoken English to written English, the likelihood of the technology misinterpreting the words spoken or dropping vital punctuation remains high. As such, an additional round of proof is likely necessary for most professional settings and critically important if the content is intended for subtitling or voice-over use.
Speech-to-text technology is also unable to replace the need for human intervention when removing those “filler words” that are so unnecessary for subtitling/voice-overs, such as “um” and “er.” Likewise, for certain languages, it could take longer to “clean up” the transcribed text than it would take for an expert to produce a transcript from scratch.
There are plenty of ways that speech-to-text programs can play a vital role in saving you and your business time; for example, by providing a quick, workable first pass where you might have had to rely on audio typists in the past. However, such technologies are not (yet) a reliable solution for every transcription task.
To find out more about how to use speech-to-text technology to complement your subtitling and voice-over projects, contact us today.