Quick and Easy Speech Transcription with Cloud-Based ML

If you listen to video presentations or podcasts, you’ll probably agree that it’s a diverting pleasure, but it’s not a particularly efficient way to extract information. For speech to be clear enough to be easily understood, broadcasters need to speak at around 150 words per minute. This means that information transferral is relatively slow – especially when you consider that the average adult can read around 250-300 words per minute, and scan large quantities of text in even less time.

That’s why adding transcriptions to your audio or video content can be a game changer: it lets your audience parse your content quickly and easily to identify areas of interest, making it more efficient for them to digest. It also unlocks the potential of your content by making it available to different audiences. If your language isn’t my language, it’s pretty easy for me to run your whole text through a translation app. Translating the audio would take significantly longer (and I might have lost interest by then…).

Once upon a time, the only way to get a reliable transcription was to sit someone down with a recording and ask them to type it up but – thank goodness – machine learning (ML) has revolutionized that painful process. Nowadays, speech transcription tasks can be performed in moments – and we’ve put together a tutorial to show you how simple and swift it can be, even if it’s something you’ve never attempted before.

The tutorial shows you how to set up speech transcription deployed as a service, running on cloud-hosted Arm servers. Essentially, you just record an audio file and send it to the server. The server uses an ML-based speech recognition service to transcribe the audio, and sends the text back to your client machine.

No knowledge of machine learning is required, and the step-by-step guide walks you through each stage of the technical deployment. (The tutorial uses Ubuntu 16.04 so some familiarity with the command-line, Linux package managers, and SSH is assumed.) You’ll need a couple of hours to set up the installations and builds, but once you’ve done that, the service will be up and running very quickly – and all on a computing cost of around $3. And with 96 cores available on a Cavium Thunder X server such as the one used here, 24 hours of English or Mandarin speech can be transcribed with state-of-the-art accuracy for just $0.50!