We recently updated premium voiceover generation to improve quality and consistency by generating voiceover using the full context of a project's script. It produces vastly improved results but might require a change in your workflow to make sure you're keeping tabs on your voiceover quota.

This update eliminates inconsistencies in:

When you apply a Premium voice, you can now expect consistent narration without encountering the problems above.

Yes. Although you can still apply your premium voiceovers as before, we recommend that you finalize your script before applying a premium voice to preserve your quota. If you make any changes to your script after applying a voice, the voiceover for the entire project will need to be regenerated, and the full duration of the narration will be deducted from your quota.

Once you're ready, you'll still select Apply to generate the voiceover, and then once it is generated, it will be applied to your project.

If you do have to make changes after applying a voiceover, you can preserve your quota by making all of the changes before regenerating.

🖥️ If you’re interested in understanding how we came to this decision, here’s a peek behind the curtain:

When we decided to partner with ElevenLabs, we wanted to make our quota calculations as cost-effective as possible for our users. We did something that even ElevenLabs doesn't do — allowed users to re-generate only the altered sections of their audio.

To achieve this, we generated voiceover in sections rather than as a single unit so that only a portion of the voiceover would need to be regenerated again.

To generate audio in parts, we sent each scene as a separate script to ElevenLabs, and that caused problems. ElevenLabs is an AI tool (a fabulous one), and like any AI tool, it needs context to generate results. Sending each scene individually meant that ElevenLabs' model didn’t have enough context to generate the consistent and realistic voiceover our users needed.

To solve this problem, we're giving ElevenLabs what it needs to do its best work. We're sending users’ entire script to ElevenLabs so that it has enough context to maintain consistency in the pitch, speed, volume, and pronunciation. We know that this might require a change to your workflow, but we're confident that the boosted realism will outweigh the learning curve.