Descript offers a cutting-edge automatic (AI-powered) transcription service, delivering near-instant, highly accurate text from audio and video. With support for 25 languages and advanced features like AI-powered speaker labeling, it meets diverse transcription needs. The service ensures data confidentiality and offers both a free plan and affordable paid options.

Transcribing files

There are a few ways you can automatically transcribe a file in Descript:

• You can transcribe files by adding them to your script; drag and drop them into the Script editor, or insert an already added file from your Project panel into the script.
• If you have imported a file directly into your project, you can transcribe it by selecting the options menu ( icon) to the right of the file’s name and choosing Transcribe file.
• You can transcribe files when creating a sequence or adding files to a sequence from the sequence editor.

When a transcription is still loading, you may notice dots appearing in place of text. These dots indicate that the transcription process is ongoing. The time it takes for the text to fully appear depends on factors like media quality, file size, and internet speed. For the best experience, ensure a stable internet connection. If the dots remain for an extended period, you may need to request the transcription to complete the process manually.

File size limitations

While there is no strict limit (in bytes) on the size of files that can be automatically transcribed, any file 15 hours or longer will automatically fail transcription.

Additionally it is important to consider the potential impact on app performance. Large files may take longer to upload and transcribe, which could cause delays or other performance issues within your project.

Split large files into smaller segments

Large files (especially those longer than 15 hours) should be split into smaller segments before uploading them for transcription. This can also help to avoid performance issues and ensure a faster transcription delivery time.

Transcription accuracy

Like most other automatic transcription services, Descript uses a third-party transcription engine for automatic transcriptions. These companies can provide much more accurate transcriptions than anything built independently.

But don’t worry; we regularly measure the accuracy of all the primary transcription services to be sure we're always providing the highest quality to our customers. We sometimes publish the results of these tests -  you can read more about these findings here.

The accuracy of automatic transcription varies significantly depending on the characteristics of the audio. With well-recorded audio and clear speech, Descript's automatic transcription can be up to 95% accurate. However, in cases of poor audio quality or difficult-to-understand speech, the accuracy can drop.

Choosing a transcription style

By default, Descript automatically selects the transcription style we believe offers the best all-around performance for most types of content.

If you'd prefer to select your transcription provider, you can change this at any time from your App Settings. Each style is powered by a different AI model with its own strengths—some are faster, some handle punctuation more precisely, and others may perform better with non-English speech or proper nouns. You can hover over the info icon in the app to learn more about each one.

How transcription time is calculated when uploading files

No matter what plan you are on, you have a monthly amount of transcription hours in Descript. When you upload and transcribe a file, the transcription time used is based on the entire file length, not just the parts with speech. This is important to keep in mind, especially when:

• Uploading multiple speaker files from the same interview or recording session. Each file will be deducted from your monthly transcription hours.
• You transcribe a file with large sections without speech. Again, the total length of the transcribed file will be deducted from your monthly transcription hours, not just the parts with people speaking.