Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.deliah.ai/llms.txt

Use this file to discover all available pages before exploring further.

If you’re new to Deliah or getting ready to set up your Voice Clone, the questions below cover everything from what you need to record to how messages reach your fans. If something isn’t answered here, contact Deliah support directly.
The minimum to train a Voice Clone is 30 seconds of clean audio. However, a longer session gives the model more to work with and produces noticeably better results. The recommended range is 1 to 3 minutes of clear, natural speech. More audio — especially recordings that cover different emotional tones and speaking styles — improves the quality and expressiveness of your clone. If you have existing recordings, you don’t need to start from scratch.
A professional condenser or dynamic microphone connected to an audio interface will give you the best results — clean signal, minimal background noise, and consistent volume. That said, a modern smartphone records at a quality level that works well for Voice Clone training, provided your recording environment is quiet. The most important factors are minimizing background noise (fans, traffic, air conditioning) and recording in a treated or soft-surfaced space to reduce echo. A walk-in closet lined with clothing is a surprisingly effective makeshift recording booth.
Realism comes from the quality and variety of your input recordings. The model learns your voice from what you give it, so the more natural, expressive, and varied your recordings are, the more convincing your clone will be. Three factors matter most:
  • Audio clarity: Clean recordings with no background noise, clipping, or reverb give the model an accurate representation of your voice.
  • Natural speech: Reading from a script in a stiff, careful way produces a different model than speaking naturally and conversationally. Aim for the latter.
  • Emotional variation: Recording across multiple emotional registers — Normal, Whisper, and Ecstasy — gives your clone more range and allows chatters to match messages to different moments and relationship stages.
Yes. You can submit existing videos, audio files, and real voice messages you’ve already sent to fans. Deliah’s team will extract the usable audio from your submissions. This is a good way to build up your training dataset quickly if you have a library of existing content. Keep in mind that the quality of the extracted audio still matters — heavily compressed video, background music, or ambient noise will reduce the effectiveness of the training data compared to purpose-recorded clean audio.
Processing time varies depending on the volume of recordings submitted and current demand on Deliah’s systems. Deliah does not publish a fixed timeline because it depends on factors outside your control. For an accurate estimate based on your specific submission, contact Deliah support directly. They can give you a realistic timeframe based on what you’ve submitted or plan to submit.
Chatters send messages using your Voice Clone. A chatter is either a member of your own team or a member of Deliah’s team, depending on how your account is set up. Chatters are the people who manage your fan conversations — they write or adapt the message content, choose the appropriate emotional variation, generate the voice audio through your clone, and send it to the fan. You set the parameters for how your clone is used; chatters operate within those boundaries on your behalf.
Your Voice Clone is designed to sound realistic and natural — fans typically cannot distinguish a well-trained clone from a message you recorded yourself. How you present voice messages to your fans is entirely your choice. Some creators are transparent with their audience about using AI tooling; others treat the messages as straightforward personal communications. Deliah does not dictate your approach. What matters most is that the messages feel genuine in tone and content, which is why recording quality and chatter personalization both matter.
The most effective way to improve your Voice Clone is to submit more recordings — especially in areas where the current output feels weak. If your clone sounds flat, record more expressive, natural speech. If the Whisper variation feels unconvincing, record more Whisper-register audio. Cleaner recordings also help: if your existing training data included background noise or compression artifacts, new clean recordings will improve the model. Contact Deliah support to discuss your specific quality concerns and get guidance on what type of additional recordings will have the most impact.
Yes. Your Voice Clone is not a fixed, one-time snapshot — it can be improved and expanded by submitting additional recordings. Updating your clone is useful if your voice has changed, if you want to add or strengthen a particular emotional variation, or if you simply want higher overall quality. Submitting more audio regularly, especially as you generate new content, keeps your clone current and expressive. Think of it as an asset you maintain and improve over time rather than something you set up once.
Your Voice Clone supports three emotional variations, each serving a different purpose in fan engagement:
  • Normal: Your warm, natural speaking voice — conversational and authentic. This is your baseline and the most versatile variation, suitable for the majority of fan messages.
  • Whisper: A quieter, more intimate register — closer, more private, and more emotionally tender. This variation is particularly effective for deeper fan relationships and premium content.
  • Ecstasy: Your most expressive and passionate register — intense, emotionally heightened, and distinctly personal. This variation is used for high-value interactions and fans at the deepest levels of engagement.
Recording all three gives chatters the full range of tools to match messages to the right moment. A clone trained only on Normal-register audio will feel limited compared to one that covers all three variations.