FAQ & Troubleshooting

Common questions and solutions

Frequently Asked Questions

Why is the output different every time?

The AI is nondeterministic. Like a human actor, it gives a slightly different performance each time. Use the Temperature slider to control this variability:

• Lower temperature: More consistent, predictable outputs
• Higher temperature: More variation and expressiveness

Why does voice selection matter so much?

Voice selection has a tremendous effect on the output. The AI inherits characteristics from the selected voice:

• If you want an introspective voice, select a voice that sounds introspective
• If you want an energetic voice, select one cloned from energetic samples
• If you want a specific dialect, choose a voice trained on that dialect

How are credits calculated?

Credits are calculated based on the number of characters in your text and the normalization mode selected:

• Basic Normalization: x1 credits
• AI-Enhanced Normalization: x2 credits

What audio format is generated?

All generated audio is delivered in MP3 format at 128 or 192 kbps (44.1 kHz) for optimal quality and compatibility.

Is there a character limit?

Mode	Character Limit	Approx. Duration
Emotions & Dialects Enabled	3,000 characters	~3 minutes
Standard Mode	10,000 characters	~10 minutes
Text to Audio (Quick)	1,500 characters	~1.5 minutes

For longer content, use the Text to Audio Studio.

Can I use the generated audio commercially?

Yes, all audio generated with your account can be used for commercial purposes according to your subscription terms.

---

Troubleshooting

Audio sounds robotic

• Try increasing the Temperature setting for more natural variation
• Increase the Expressiveness slider slightly
• Enable Emotions & Dialects Mode for expressive content
• Try a different voice that matches your content style

Output is inconsistent or unstable

• Lower the Temperature setting
• Lower the Expressiveness slider
• Use longer prompts (250+ characters recommended for Emotions mode)
• Check that your source audio (for voice cloning) is clean and noise-free

Generation is slow

• Longer texts take more time to process
• AI-Enhanced normalization requires additional processing
• Check your internet connection

Unexpected pronunciation

• Use AI-Enhanced Normalization for Arabic text (adds diacritics automatically)
• Add manual diacritics (تشكيل) to ambiguous words
• Write numbers as words instead of digits
• Expand abbreviations (e.g., "Dr." → "Doctor")
• Try a different voice - some handle certain words better

Dialect not working correctly

• Ensure the selected voice matches the dialect in your text
• Use Basic Normalization for dialect content (not AI-Enhanced)
• Start your text with a strong dialect word to "prime" the model
• Enable Emotions & Dialects Mode

Voice clone doesn't sound like the original

• Increase the Similarity slider
• Ensure your training audio was high quality (no noise, echo, or reverb)
• Check that training audio had consistent tone and energy throughout
• If source had noise, lower the Similarity to reduce artifacts

---