vocalove

Qwen3 · zero-shot voice cloning

Qwen3 TTS Clone Voice

Qwen3 TTS Clone Voice turns a reference recording into a reusable speaker embedding, then drives Qwen3 text-to-speech so new lines sound like the same person — zero-shot, no per-voice fine-tune.

Qwen3 TTS is open source with official docs and public weights on Hugging Face — On Vocalove you never call the API yourself. Pick Clone voice in the tool above.

Create your story

Voice
Script
Voice model
0 / 1000

Free to preview in your browser — no credit card. New accounts start with 5 free credits that never expire, plus a free daily pool. Watermark-free HD exports use credits only when you're ready to download.

What is Qwen3 TTS Clone Voice?

Qwen3 TTS Clone Voice is Alibaba’s Qwen3 speech family built for zero-shot voice cloning. Upload a clean reference clip — even a few seconds — and the model extracts a speaker embedding: a compact voice fingerprint you can reuse to synthesize many new scripts without re-uploading the sample every time.

Unlike one-shot models that pair reference audio with each generation, Qwen3’s clone path is two steps: first create the embedding, then pass it to Qwen3 text-to-speech with your new text. Optional reference transcripts can improve fidelity when the words in the sample are known.

The broader Qwen3-TTS family also advertises voice design — describe a brand-new voice in text without any reference clip. That is a separate mode from clone-voice; this page and the Vocalove tool above focus on cloning from a sample you upload.

The two-step pipeline — embedding first, then TTS — is especially useful when you will synthesize many scripts from the same speaker without re-uploading the reference each time.

  • Zero-shot cloning

    No custom training run — a short reference clip is enough to capture timbre and speaking style.

  • Reusable embedding

    Generate a speaker embedding once, then synthesize many scripts from the same voice fingerprint.

  • 0.6B & 1.7B sizes

    Pick the smaller model for speed or 1.7B when quality matters more than latency.

  • Pairs with Qwen3 TTS

    Clone Voice outputs an embedding file; Qwen3 text-to-speech consumes it for final audio.

How Qwen3 TTS Clone Voice works

Reference audio becomes a speaker embedding; new text becomes speech through the matching Qwen3 TTS model.

  1. Step 1

    Upload reference audio

    Provide a clip of one speaker — up to about five minutes, though a short clean sample usually works best. MP3, WAV, and similar formats are accepted.

  2. Step 2

    Create speaker embedding

    The clone-voice model returns a safetensors speaker embedding — a reusable voice fingerprint. Optionally include reference_text (the transcript of the sample) to improve later synthesis quality.

  3. Step 3

    Synthesize with Qwen3 TTS

    Pass the embedding URL to Qwen3 text-to-speech with your new script. The TTS model ignores preset voices when an embedding is supplied and speaks in the cloned timbre instead.

Qwen3 TTS Clone Voice at a glance

Key inputs and outputs on Vocalove — no API setup required.

Model sizes
0.6B (faster) or 1.7B (higher quality) — use the same size for embedding and synthesis
Pipeline
Two-step: speaker embedding from reference audio, then Qwen3 text-to-speech
Reference audio
One file, up to ~300 seconds; short clean speech recommended
Optional input
reference_text — transcript of the sample for better quality
Clone output
speaker_embedding — safetensors voice fingerprint file
TTS output
Synthesized speech audio from your script

What Qwen3 TTS Clone Voice is good for

When you need a specific timbre and may generate many lines from the same sample.

  • Memorial & tribute messages

    Capture a loved one’s voice once, then produce multiple lines of comfort or remembrance without re-processing the original recording each time.

  • Multi-line keepsakes

    Create an embedding from a voicemail, then iterate on wording — birthday wishes, holiday messages, or reunion notes in the same voice.

  • Talking photo audio

    Generate a cloned speech track, then pair it with a portrait video so the face lip-syncs to words that sound familiar.

  • API & batch workflows

    Reuse one embedding across many TTS calls — efficient when a product generates lots of speech from the same speaker profile.

Qwen3 Clone Voice vs other cloning approaches

How Qwen3’s embedding pipeline compares to one-step cloners and preset narration.

 Qwen3 Clone VoiceOne-step clone (e.g. F5-TTS)Preset narrator
ArchitectureTwo-step: embedding, then TTSOne-step: reference + text togetherPreset narrator, no sample
Reuse same voiceEmbedding file — synthesize many scriptsRe-upload sample each generationPick a fixed preset voice
Sounds like someone you knowYes — zero-shot cloneYes — zero-shot cloneNo — generic narrator
BillingVaries by host APITypically per character on hosted APIsPer character or per request on host

Voice cloning on Vocalove

Use the tool above to record or upload a sample, type your script, and download cloned speech in the browser. Qwen3 TTS Clone Voice on API hosts follows a two-step embedding + TTS pipeline — useful when you need to reuse the same voice across many generations.

Try voice cloning on Vocalove in three steps

Upload a sample, type your script, and generate cloned speech in the browser.

  1. Step 1

    Record or upload a sample

    Three to thirty seconds of clear speech in a quiet room. mp3, wav, or m4a work; you can also record in the browser.

  2. Step 2

    Type your script

    Write what you want heard. English and Chinese both work well; natural phrasing beats long run-on sentences.

  3. Step 3

    Generate cloned speech

    Preview and download audio — usually within one to three minutes. Continue to Talking Photo Videos when you want lip-sync.

Jump to the speech tool ↑

Why Qwen3 TTS Clone Voice?

If you only need one line, any zero-shot cloner works. Qwen3 stands out when you will synthesize many scripts from the same speaker — the embedding step amortizes the reference audio across unlimited TTS calls.

The 0.6B models suit lighter workloads; 1.7B targets higher-quality cloning and synthesis. Pair clone-voice and text-to-speech at the same size for best results.

  • Optional reference_text helps the model align timbre with what was actually said in the sample.
  • Reference clips up to five minutes are accepted — still, shorter clean speech beats long noisy recordings.
  • Pair generated audio with a portrait on Talking Photo Videos when the script is ready for lip-sync.

What to do after cloned speech

Happy with the audio? Download the MP3. Ready for lip-sync? Open Talking Photo Videos.

Make a talking photo videoBack to the speech tool

Qwen3 TTS Clone Voice FAQ

Is Qwen TTS the same as Qwen3 TTS?
Qwen TTS is the common name for Alibaba's speech line; Qwen3 TTS is the current generation. On Vocalove this page covers Qwen3 TTS Clone Voice — searches for "Qwen TTS" land on the same clone-from-sample workflow.
What is Qwen3 TTS Clone Voice?
A zero-shot voice cloning model in the Qwen3 speech family. It reads a reference audio clip and outputs a speaker embedding — a reusable voice fingerprint — that Qwen3 text-to-speech uses to synthesize new scripts in a similar timbre.
What is a speaker embedding?
A compact file (safetensors format) that encodes voice characteristics extracted from your reference clip. Once created, you pass its URL to Qwen3 TTS so each new script reuses the same cloned voice without re-uploading the sample.
0.6B vs 1.7B — which should I use?
0.6B is lighter and faster; 1.7B generally gives higher-quality cloning and synthesis. Pick the same size for both the embedding step and synthesis — 1.7B with 1.7B, for example.
What is reference_text for?
An optional transcript of the words spoken in your reference audio. Providing it can improve synthesis quality when the clone embedding is used later, because the model knows what was said in the sample.
How long should my reference sample be?
Up to about five minutes is accepted, but three to thirty seconds of clear single-speaker speech is the practical sweet spot.
Does Qwen3 TTS Clone Voice make talking photo videos?
No — it outputs speech audio (via the TTS step). Talking photo videos on Vocalove add a portrait so the face lip-syncs. Generate audio first, then open Talking Photo Videos.