Voice Cloning
Clone a voice from reference audio into a reusable private speaker, then use its speaker ID anywhere a ListenHub voice is expected.
Turn one to six reference recordings into a private speaker with its own ID. Once confirmed, that
ID works everywhere a ListenHub voice is expected — /tts, /podcast, the explainer and slides
skills, and the API.
Trigger
Invoke this skill with /voice-clone, or use any of these phrases:
| Phrase | Language |
|---|---|
clone my voice / voice clone / custom voice | English |
克隆我的声音 / 克隆音色 / 声音克隆 | Chinese |
用我的声音 / 自定义音色 / 复刻声音 | Chinese |
Requires ListenHub Skills to be installed — see Getting Started.
Quick Example
Clone my voice from ./recording.m4a and read this with it: "Welcome back to the show."The agent clones, plays you a preview, and asks before keeping the voice — confirming is the step that can cost credits.
Which cloning do you want?
Two skills clone a voice, and they produce different things.
/voice-clone | /listenhub-voice | |
|---|---|---|
| Produces | A persistent private speaker with an ID | Audio for this one request |
| Stored | Yes — occupies a voice slot on your plan | Nothing is kept |
| Reusable | In every skill and API call that takes a speaker ID | No |
| Costs | Free while your plan's per-period confirmations last, then 300 credits per confirmation | Nothing beyond the generation itself |
| Use it when | You will speak in this voice again | You need one clip and nothing more |
How a clone runs
Supply the reference audio
One to six local files — .mp3, .wav, .m4a, .flac, .ogg, .aac. Clean speech with no
background music clones best. A single file is the common case.
Confirm consent and language
If the recording is not your own voice, the skill asks whether you have the speaker's permission,
and stops if you do not. Then it asks which language the recording is in — zh or en on the
logged-in path. The API-key path reaches more languages; see the API reference.
Clone and preview
The clone runs as a task and returns a preview of a temporary voice. Nothing is stored and nothing is charged at this point.
Keep it — or don't
Only when you say to keep it does the skill name the voice and confirm it into a permanent speaker. Unconfirmed tasks expire after 7 days.
Use the speaker ID
The skill reads back the new speakerInnerId and hands it to you. Pass it anywhere a voice is
expected.
Confirming is the step that costs. It is free within your plan's per-period quota, and 300 credits each after that. The skill never confirms on its own initiative, and never authorizes a credit charge without an explicit yes from you.
Limits
| Item | Value |
|---|---|
| Reference files | 1–6 per clone |
| File size | ≤5MB per file through this skill, ≤20MB in total (the API itself accepts 20MB per file) |
| Plan | Voice cloning requires a paid plan — a free account gets an upgrade error at the confirm step |
| Voice slots | Your plan caps how many cloned voices you may keep at once |
At the slot cap, delete a voice to free one. That frees the slot but does not refund the confirmation already spent on it.
Managing cloned voices
The skill can list your voices with the plan's slot cap and remaining confirmations, inspect one, rename it or change its gender, and delete one to free a slot. Ask for what you want in plain language — "list my cloned voices", "rename this voice", "delete the old one".
Cloned voices are private to your account and appear alongside the official catalogue when you list speakers.
Consent
You must have the speaker's permission before cloning their voice. If the recording is not your own, the skill asks whether you have that permission and stops if you say no. It has no override, and it does not verify the answer — the responsibility is yours.
API Reference
See the Voice Cloning API reference for the endpoints, the task lifecycle, the confirmation charge, and the speaker-management calls.