Features

Limecraft Transcriber gets a complete makeover

Maarten Verwaest
September 5, 2016

As part of its latest release of Limecraft Flow, Limecraft launched a total makeover of its Transcriber application.

 

Screen Shot 2016-09-05 at 23.30.34

 

Transcriber is an application for “Computer Assisted Logging” or “AI Transcription“, whereby as a user you can tune, edit and correct the automatically generated transcript. Used by journalists and archivists, AI Transcription enables them to save up to 80% of the time to turn audio into text.

Using Automatic Speech Recognition (ASR), you can now turn interviews, speeches, news conferences, and meetings in timed text in minutes. With an accuracy of 95% or more the result is already good enough to pick and markup the quotes you intend to use for editing or publication. For these fragments where you require a 100% accurate transcription, Transcriber enables you to efficiently edit and correct the transcript. When finished, you can export your transcripts or your list of quotes in your format of choice.

As a side effect of the transcription process, because each word is now time-stamped, your media have become more easily searchable and much more accessible. Using Transcriber, you’ll never loose that valuable quote anymore.

The need for Transcription

Whether you are an occasional podcaster, a journalist, a documentary maker, a subtitle operator or an archivist, chances are you spend a considerable part of your professional career in transcribing interviews or in describing the contents of your audiovisual material as accurately as possible. You do so because pixels and the waveforms on their own are hard to get. So to make sure you can find the needle in the haystack, or more specifically to pick that particular quote, you rely on an accurate transcription.  Strange enough, while speech technology has been out there since 20 years, transcription is still done manually in most cases. To crack that challenge, we had first of all to find out why.

Unfortunately, despite many technological innovations in speech processing in the last couple of years, the result of machine transcription is often not good enough. Hence, media professionals find themselves transcribing the content of interviews, speeches, using a conventional text editor.  Manual transcription takes 4 to 8 hours of work per hour of audio. All together, every day in our life, an incredible volume of audiovisual content is manually processed by journalists, archivists and media professionals of all kinds.

How Limecraft raised the Bar 

At Limecraft we investigated why professionals don’t like the result of off the shelf speech technology and we found a way bridge the gap between state of the art technology and the requirements of the professional user. To that purpose, we improved the baseline performance of the underlying ASR technology, we wrapped it in a slick application used by the professional to control and modify the transcription and eventually we added a self-learning algorithm, so the speech engine is smart enough to pick up new words as users correct the machine generated transcriptions.

We found that the biggest complaints about the use of commercial Automatic Speech Recognition engines boiled down to their poor baseline performance. To achieve an accuracy of 90% or more in real life circumstances, we had to reduce background noise and train the system for real speakers and the use of dialects. This, in combination with a broad large lexicon (database of words), enables you to use ASR technology with unprecedented accuracy.

The next improvement was the Transcriber application on its own. While Transcriber shows the machine generated transcript as if it were a conventional text editor, because every word is timestamped and synchronised with a player window, as a user user you can easily navigate through the transcript. By doing so, you can efficiently identify and edit the specific sections that require improved accuracy. Subsequently you can download the transcript in PDF, excel or subtitle formats.

However, the most effective improvement is the self-learning algorithm. As you correct the transcripts, Limecraft understands which should have been part of the lexicon and it will update the ASR engine in the background. After a couple of instances where you correct the transcript, the engine will spontaneously start recognising new concepts as they come along.

How and why other people are using Transcriber

There are countless applications of the transcription application and for the sake of illustration we will highlight a couple of use cases of how Transcriber is currently used in professional environments.

  • Documentary makers usually start by more or less accurately transcribing large volumes of live action or archive stock, often 100 hours of raw material per 50’ episode. By using Transcriber, they save 80% of the offline edit budget, they significantly accelerate the proces and they increase the chances to find the right fragment.
  • Scandinavian and Belgian broadcasters systematically subtitle all of their content. By exporting the machine generated transcript and importing that in their specialised subtitle application, they directly cut the cost of producing subtitles by 50%.
  • A large community of podcasters and journalists use Limecraft Transcriber every day to turn their audio-only interview into a text. Using the speech technology, they reduce the time required to process the audio by about 50%.
  • A dutch company specialised in capturing and processing the minutes of official meetings uses Limecraft to generate full transcripts of the meetings. By comparing with the original agenda, they automatically generate the as run agenda. Without Limecraft, such application would not have been possible.
  • Several archive service providers, i.e. companies that care for the archiving process on behalf of their customers, use Limecraft Transcriber to improve the commercial exploitation of the available assets because browsing through and selecting accurately transcribed content is a lot more efficient compared with searching in a database of poorly logged content.
  • You want to give it a try, now what?

💡 If you would like to give it try, we warmly welcome you and we are happy to help you onboarding. Please note AI transcription is available in the free plan, albeit as a paid-for option.