Voice Enrollment
Distinguish enrollment, Voice ID, and actual spoken audio.
Get permission for the specific use
Voice enrollment can reproduce identifying vocal traits. Ask the voice owner for explicit authorization for the recording, synthesis, intended audience, and retention before uploading anything.
- Agree on a script and distribution scope with the speaker; never upload a scraped interview to impersonate someone.
- Keep consent documentation outside public prompts and confirm the node’s authorization checkbox truthfully.
Capture a usable reference recording
Use a clean recording with one speaker and consistent microphone distance. A music bed or overlapping speaker can become part of the learned sound or cause enrollment failure.
- Trim long silence, confirm the correct speaker, and audition the full source before selecting a file.
- If using HKGAI’s optional reference-voice panel, follow its displayed 1–30-second, at-most-10-MB guidance and provide an exact transcript when helpful.
Create a Qwen voice ID
The Audio → Voice Cloning node registers a Qwen voice through the server. A successful enrollment yields an ID, not automatically a finished narration for a separate script.
- Place Voice Cloning, select the authorized reference file, set a simple Preferred name and Language, and check the consent box.
- Press Create voice; wait for the returned voice identifier and check whether the node reports a fallback mode.
Turn enrollment into approved speech
Voice Cloning can generate speech from its own Speech text; a separate Qwen Voice TTS node can reuse the Voice ID and upstream text. Both produce audio only after synthesis.
- Enter a short authorized line and press Generate speech on the cloning node, or connect its Voice output to Qwen Voice TTS.
- Compare the generated words with the approved script; avoid using an unexpectedly similar but unapproved speaker identity.
Handle missing and retired voices
Qwen voice listing and generation depend on DashScope access. An old ID, a fallback registration, or a temporary audio URL calls for different remedies; do not paste API keys into prompts.
- If a cloned voice does not appear in Qwen Voice TTS, refresh its voice selector and verify the enrollment node completed successfully.
- If consent is withdrawn, use Delete selected for the registered voice and ask the administrator about any retained assets or provider records.