ETD MDL AI Voice Tool

🎨 Design a new voice

Describe the voice you want (age, gender, tone, pace, emotion) and Qwen3 will create it. No reference audio needed.

Note: uses the Qwen3 1.7B VoiceDesign model. Keep the text short (a sentence or two) to audition a voice, then save it to the library and use it in the other tabs for longer scripts.

Voice description

Quick presets:

Text to speak Language The text above should be written in this language

⚙️ Advanced Options

Random seed (0 for random; batches use seed, seed+1, …)

Batch size Generates several takes in one job, each with a different seed, so you can pick the best.

💾 Save to voice library

Take to save Voice Name

Gender Male ♂️ Female ♀️

⚡ Turbo Mode

Available Paralinguistic Tags: [clear throat] [sigh] [shush] [cough] [groan] [sniff] [gasp] [chuckle] [laugh]

Example: "Hi there, I'm Bouncy [chuckle], someone said hungry? [gasp]"

Text to synthesize (unlimited length - smart chunking enabled)

Quick Insert Tags:

Reference Voice Voice library Upload / record

Select Voice (Required for Turbo) Turbo requires a reference voice clip for cloning

Language: English only. Note: Turbo model requires a voice reference.

⚙️ Advanced Options

Random seed (0 for random; batches use seed, seed+1, …)

Batch size Generates several takes in one job, each with a different seed, so you can pick the best.

💡 Tips for Best Results:

Use paralinguistic tags naturally in your text
Tags work best mid-sentence or at natural pauses
Combine multiple tags for complex emotions
Generation is ~3x faster than standard model

Convert speech with a trained RVC character voice

Convert one recording or a batch of files with an RVC voice model. Output is always WAV.

No RVC models found on the server. Add .pth files to the rvc_models folder and restart it.

AI Voice (RVC model)

Input Single file Multiple files

Input Audio

Transpose (semitones) Li Yen: 6 (try 4/5 if audio is glitchy) · Kevin (Poodle): 12 to 15. Unlike the other tabs, speed stays the same.

Pitch Extraction Algorithm pm harvest crepe rmvpe 'rmvpe': best all-round, recommended default · 'pm': quickest, slightly rougher speech · 'crepe': smooth, stable pitch · 'harvest': good for deep or low voices. All take a few seconds per line; on long recordings crepe and harvest take a little longer.

Median Filtering Harvest only: 3 or more applies median filtering to the pitch, which can reduce breathiness. Search Feature Ratio Higher = more like the target voice but can sound artificial; lower = more natural but less like the target.

⚙️ Advanced Options

Resample Output Audio (Hz) Post-process to this sample rate. 0 = keep the model's rate. Use Volume Envelope The closer to 1, the more the output's own loudness envelope is kept; lower follows the input's loudness. Protect Voiceless Consonants and Breath Sounds Prevents tearing artifacts. 0.5 disables; lower protects more but may reduce index accuracy.

Clone any voice instantly!

Upload or record a clear audio sample (5-30 seconds)
Name your voice and select gender
Select the language
Click "Clone Voice" and use it in any tab!

Tips: clear, high-quality audio; no background noise; 10-20 seconds is ideal; multiple sentences work better.

Voice Name

Gender Male ♂️ Female ♀️

Voice Language Select the language of the voice sample

Reference Audio Sample

Your Cloned Voices

Voices are stored in the voice_samples folder and appear in every tab once saved (on the website, after the server has synced them, usually a few seconds).

Your recent jobs (newest first). Results stay in the shared folder under outputs/.

😆 ETD MDL AI Voice Tool 😀

Can't reach the AI Voice Tool service

🎨 Design a new voice

⚡ Turbo Mode

💡 Tips for Best Results:

Convert any voice to another!

Bulk convert audio files to another voice!

Convert speech with a trained RVC character voice

Clone any voice instantly!

Your Cloned Voices