Limecraft is used by many newsdesks and documentary producers for audio transcription. Using Automatic Speech Recognition or AI transcription, the bulk of the work is automated. Yet it is an automated process, thus prone to errors. In cases where you need a 100% accurate transcription, you can use Limecraft for post-editing the transcript. This article explains the best approach to correct a automatically generated transcription in the shortest possible timeframe.

As we explained in another article, you can use Limecraft to automatically transcribe audio or video in minutes. However, turning naturally spoken language into a correct transcript is hard for machines, so you will notice occasional errors in the recognition of words, speaker segmentation and interpunction. As a matter of fact, the less correct the language used by speaker, the more errors you will notice. 

So while the machine generated transcription might be good enough for certain purposes like indexing or pre-cutting a story, you might need a 100% accurate transcript for others like compliance reasons, publication or preparing subtitles. In these cases you can review and edit the transcript.

We noticed that novice users intuitively post-edit the transcription from the top down, correcting all types of errors in a single pass. They will spend 4 to 8 minutes of post-editing depending on the quality of the audio. Using a more efficient strategy, you should work step by step from the top down. This method, developed by experienced users, will enable you to reduce the time for post editing by 50%. This article describes the different steps of this method.


Overview of the 4-step Method for efficient post-editing of AI Transcription

The key to efficiently adjust the automatically generated transcript is to work from the outside inwards. In a first pass, you correct the paragraphs by approving or adjusting the speaker segmentation. The first review can go faster than real time, and it will give you a sense of which areas need further attention. In a second pass, you add commas and full stops as appropriate. At this point the skeleton of the transcript is ready and you just need to replace or add individual words or, in case of poor recognition, replace or add entire sentences.

4 step strategy for efficiently post-editing AI transcripts

Step 0 - Before you begin

We assume you have a valid subscription, your account was provisioned with transcription credits, you were able to upload a file and you successfully transcribed a clip.

To enable editing, toggle the transcriber app from "VIEW" to "EDIT" mode. Now you can correct the transcription, including segmentation, speakers, punctuation and words online.

Visualisation of the result of Automatic Speech Recognition (ASR) as part of your Limecraft Workspace

Step 1 - Speaker segmentation

In a first pass, we advise to adjust the speaker segmentation as appropriate. It is easy to execute faster than real-time, by increasing the play out speed.

Screenshot on how to correct speaker changed in Limecraft

To cut a paragraph, put the cursor in the appropriate spot and hit ENTER. To remove a line break, use BACKSPACE.

Step 2 - Interpunction

As a second operation, we recommend to add or remove commas and full stops as appropriate. These are critical for managing the timing of the transcription, because punctuation marks are used as time-anchors in many operations later on, such as subtitling.

Correcting the punctuation is a process that can also be performed faster than real time, by hopping forward in the transcript using CONTROL+Click (windows) or CMD+Click (Apple).

Step 3 - Correcting word errors

When the segmentation and the interpunction are approved, your timing framework is fixed. Now you can safely replace individual words or groups of words without loosing the timing information. 

Tip: you can use the confidence scores to highlight poorly recognised sections to give you a visual overview of the areas that require your attention.

Screenshot of how to highlight confidence scores of AI transcription in your Limecraft Workspace

Important note: when replacing groups of words, it is critical to correct fragment per fragment (between the punctuation marks).

Step 4 - Adding non-recognised segments

As a last step, a step which only applies in cases of audio which is difficult to transcribe and where the Automatic Speech Recognition engine doesn't detect anything meaningful, you may need to replace or add entire fragments by hand.

To start, make sure the playhead is at the beginning of the section and hit SET IN. At the end of the fragment, use SET OUT to pause the player. The transcriber will display a pop up that allows you to replay the section as appropriate, while adding or replacing the text. When done, use CONFIRM to acknowledge the change.

Screenshot of how to add non-recognised segments when reviewing AI transcripts in your Limecraft Workspace