> For the complete documentation index, see [llms.txt](https://learn.firecut.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://learn.firecut.ai/features/remove-repetition.md).

# Remove repetition

## ❓ What is it?

Certain types of video tend to have a lot of repetition in the raw footage (e.g., single speaker setup where they are recording a script, doing multiple "takes" of certain phrases).

**Remove repetition** detects when the speaker is repeating phrases in the video, and allows you to select one "take" to keep while deleting the others.

## 🤔 How does it work?

Remove repetition in FireCut:

* Analyses the words being spoken in the selected portion of your sequence and uses AI to detect when phrases are being repeated
* For each phrase that is being repeated, shows you a list of all sections where you have a "take" of that phrase. The words might vary slightly between these "takes", so FireCut does its best to pick up approximate matches as well as exact matches
* Lets you preview each "take" by clicking on it
* Once you know which "take" you want to keep (this usually tends to be the last one in each set, as the speaker will probably keep going until they think they've recorded a good "take"), you can select it with the tick button. This deletes all the other "takes" in that set
* Keep going through the phrases until you've selected your favourite "take" for each, or click **Auto-accept last takes** or **Auto-accept longest takes** to do every set at once
* Adds a marker to your sequence at every cut if you turn on **Add markers to indicate cuts** above the list. Each marker is called "Repetition removed" and lists the removed takes in its comment, so you can find and check every cut later
* If you'd rather find the repetitions differently, you can switch the detection source from AI to manual matching after you get your results (see [Changing the detection source](#changing-the-detection-source) below)

{% embed url="<https://www.youtube.com/watch?v=rzhOvEYDWP0>" %}

## ⚙️ Settings

There are 2 main settings for remove repetition:

1. **Scope:** You can decide whether to run the operation on your **Full sequence** or only on a certain portion that you specify with **In / Out points** (you can place these in your sequence using the `I` and `O` hotkeys)
2. **Language:** We are constantly adding support for new languages. However, please note this is experimental and may create unpredictable results when compared with English. This is the case for every AI tool even if they don't mention it, because [training sets for AI tend to be in English](https://www.wired.com/story/chatgpt-non-english-languages-ai-revolution/). On occasion, this might cause the operation to fail -- please do let us know (<support@firecut.ai>) if this happens, so we can improve it for you!

There's no longer a choice between Basic and Advanced mode: every run uses AI to find the repetitions, and you can switch to manual matching afterwards.

## ⚒️ Changing the detection source

When your results appear, the top of the list shows **Repetition detection source: AI selection**, meaning the AI found these repetitions. To change how they're found:

1. Click **Edit in transcript view**
2. At the top, choose the source:
   * **AI SELECTED:** the repetitions the AI found
   * **MANUAL:** FireCut matches repeated phrases in your transcript using the settings below
3. Click **Preview and choose best takes** to go back to the list. It now shows the source you chose (for example **Repetition detection source: Manual**)

{% hint style="info" %}
**Edit in transcript view** is only available until you accept a take, because after that the transcript no longer matches your sequence. To change the source after accepting takes, run Remove repetition again.
{% endhint %}

### Manual matching settings

These settings fine-tune how repeated phrases are matched, and only apply when the source is **MANUAL**. Under **Presets**, click **+** to save your settings under a name, then click a preset to use those settings again. **Default** brings back the standard settings.

* **Minimum phrase size:** The minimum number of words that have to be repeated in order to be detected as a "match". The default is 8 words, and you can choose 5 to 20. Increase it to ignore short repeated phrases
* **Tolerance:** How loose of a match will be detected as a repetition. The default is 40%.
  * 0% means only exact matches will be detected
  * 80% means even loose repetitions will be detected (this might catch some "false positives", i.e. phrases that were only slightly similar)
* **Search radius:** How far away to look for a match, from 0 to 80 words (the default is 50). If you increase this, FireCut will look for repetitions even far away from your initial phrase
* **Boost multiple reps:** Lets shorter phrases count as repetitions when they're repeated more than twice. The default is 2.0x, and you can choose 1.0x to 3.0x (at 1.0x there's no boost)
* **Boost sentence start:** Increase it to prefer phrases that begin at the start of a sentence. The default is 60%, and you can choose 0% to 100%
* **Extend last take:** How FireCut should determine where exactly the last occurrence ends. Let's say we found 3 repetitions of a phrase. We know that the first 2 occurrences of the phrase end whenever the next occurrence starts. However, there's no fixed ending point like this for the last occurrence. This setting lets you customise how to define that ending point:
  * **Extend to end of sentence** (the default): The next full stop is the ending point
  * **Use the length of the longest occurrence:** This will use the longest occurrence so far, and make the assumption that the last occurrence is of that length as well. This is unlikely to be 100% exact, but you might prefer this behaviour if you know the phrases are similar lengths and don't necessarily end in a full stop
  * **Use phrase size:** This uses your **Minimum phrase size** setting above

## 💡 Tips for best outcomes

{% hint style="info" %}
Because this feature's use case is to remove repeated takes, it will detect repetition only when it is **consecutive** (i.e., the speaker has done a few takes right next to each other). If you do a second take 5 minutes after the first one, and there are other things being discussed in between, FireCut won't mark these.
{% endhint %}

* **Before selecting your favourite take**, make sure FireCut has marked the start and end of that take exactly as you want it (by clicking on the take in the list, which will mark it on your timeline)
* **After completing the process, do a double-check** to ensure your take has the right start and end points -- occasionally, FireCut might chop off starts / ends of words
* **Make sure your audio is loud and clear**, with any music / sound effects muted
* **Try it on a short section of your timeline first** to get familiar with the tool; processing larger sequences can take a while
* **Don't work in the timeline** while FireCut is working in the background

## 🙋‍♂️ Common questions, issues, watch-outs

* **"It didn't detect all the repetitions"** --> Make sure you have a clear speaker in the audio so that your script can be picked up and analysed. If there are any sound effects, multiple speakers, music, noise, etc., this can throw off the analysis in some occasions
* **"It didn't work on my non-English video"** --> Please drop a note at <support@firecut.ai>. We are keen to make sure this feature works reliably on all supported languages!
