TAKE 01 / Relaxed and conversational
Everyday voice
The rain has stopped, and the street is quiet again. I’m making a cup of tea before I head out for a walk.
THE RECORDING DESK / FREE STARTER KIT
Your voice.
A better first clone.
Start with a clean recording, not a perfect microphone. Three scripts, a practical checklist, and a clear path from your first sample to speech on your Mac.
No signup. PDF download. The kit is free; Murmur is a separate purchase.
01 / MAKE A CLEAN SAMPLE
Your recording checklist
Use these checks before importing your audio. Your recording stays with you; this page does not record or upload audio.
These checkmarks are temporary and reset when you reload. Download the kit to keep a copy.
02 / READ IT YOUR WAY
Three original recording scripts
Choose one that fits your intended delivery. Timing depends on your pace: measure the audio and use a complete-sentence excerpt for your chosen model.
TAKE 02 / Steady and easy to follow
Clear narrator
Before you begin, find a quiet place and take a comfortable breath. Read at your usual pace, leave a little space between sentences, and let your voice sound like you.
TAKE 03 / A consistent delivery over several sentences
Longer reference
This morning, I opened the window and heard a bicycle passing along the street. Somewhere nearby, someone was preparing breakfast. I put my notebook on the table and started planning the day. There was nothing unusual about the moment, but I wanted to remember it clearly. Sometimes a familiar voice can make even a simple story feel worth listening to.
Use and adapt these scripts freely. If you change a word while recording, update the transcript to match.
03 / MATCH THE SAMPLE TO THE MODEL
How long should you record?
These are Murmur’s current app recommendations, checked September 5, 2026. The selected model’s in-app guidance takes precedence as versions and runtimes change.
| Model | Recommended sample | What to know |
|---|---|---|
| Chatterbox Turbo | Aim above 5 seconds, up to 15 seconds | More than 5 seconds required; 15 seconds is guidance, not a hard cap. |
| Qwen3 Base / OmniVoice | 3–10 seconds | Exact transcript required. Murmur accepts up to 210 seconds. |
| Spark | 5–15 seconds | At least 5 seconds and an exact transcript required. |
| Fish S2 Pro | 10–30 seconds | At least 3 seconds; up to 210 seconds. Transcript needs depend on the runtime. |
Always keep an exact transcript, even when a model can run without it. Check language coverage and commercial-use terms for your selected model.
Explore models and language coverage →04 / FROM RECORDING TO RESULT
Make your first clone in Murmur
- Open Murmur on an Apple Silicon Mac running macOS 15 or later. Check the model’s memory requirements first; current optional cloning models require at least 16 GB unified memory in Murmur.
- Record one script in your normal voice. Measure the actual recording length and choose a complete-sentence excerpt that fits your model’s guidance.
- Open Clone Voice, import or record the sample, and enter its exact transcript. Give it a recognizable name and choose Save Voice Sample.
- Install and select a model that supports reference cloning. Choose your saved sample and generate the test sentence below. Kokoro provides preset voices; it does not clone your voice.
- Compare the result with your recording. Change one thing at a time: the sample, the model, or its settings. Once it sounds right, try a paragraph and export the audio.
If your first take sounds off
It doesn’t sound like me
Try a cleaner sample in your normal register. Compare a second complete-sentence excerpt with the same test text.
It sounds noisy or distorted
Listen to the original first. If it clips or contains strong room echo, make a fresh recording rather than adding more processing.
Words are missing or repeated
Check the transcript against the audio, trim long silence, and test a shorter output sentence.
The accent or language changes
Check that the selected model supports your target language. Start with reference audio and output text in the same language.
Murmur flags the sample length
Check the selected model’s guidance below. Keep a complete sentence within the supported range; don’t pad a short sample with silence.
READY TO HEAR IT?
Bring your voice to your next project.
Generate speech locally after model setup. Murmur runs on Apple Silicon with macOS 15 or later; check memory requirements for the model you want to use.