Qwen-Audio 3.0 TTS Plus for E-Commerce Voiceovers

Qwen-Audio 3.0 TTS Plus is a text-to-speech model, not an image generator — so if you came here expecting AI product photos, the tool you want is an image model like GPT Image 2. What this model is genuinely good at is turning a written script into a clean voiceover in a few seconds, which is exactly what most e-commerce teams actually need audio for: product video narration, promotional voice lines, and cross-border listing scripts that mix languages mid-sentence. Below are two real voiceovers we generated through HiAPI's /v1/tasks endpoint, plus the schema, pricing, and workflow notes for wiring it into a listing pipeline.

What e-commerce teams actually use TTS for

Most stores don't need a voice actor on retainer — they need a fast, repeatable way to narrate short-form product videos, add a voice track to promo clips for platforms like TikTok Shop or Reels, and localize listing copy for markets where the buyer expects to hear the language they shop in, not just read it. A handful of scripted seconds per SKU, generated on demand, covers all three without a recording studio.

Two real voiceovers, start to finish

We ran two real scripts through the model — not placeholder text — to see what the output actually sounds like and costs.

Test 1: English product-video voiceover

Request:

{
  "model": "qwen-audio-3.0-tts-plus",
  "input": {
    "text": "Introducing the AeroFit wireless earbuds. Twenty-eight hours of battery life, active noise cancellation, and a charging case that fits in your pocket. Free shipping ends tonight.",
    "voice": "longanlingxin"
  }
}

Real result:

{
  "taskId": "tk-hiapi-01M33PRBVZJDDXKMHK7ANDG5DK",
  "status": "success",
  "model": "qwen-audio-3.0-tts-plus",
  "created": 1790052151,
  "completed": 1790052159,
  "storage": "temp",
  "output": [
    {
      "type": "audio",
      "url": "https://temp.hiapi.ai/7c6ttvrbpt/01M33PRBVZJDDXKMHK7ANDG5DK-0.mp3",
      "artifactId": "187560",
      "expireAt": 1790656959
    }
  ]
}

That's a 178-character script, generated in 8 seconds, for about $0.007. We left format unset, so the output defaulted to MP3 — no separate encoding step needed before dropping it into a video editor's voice track.

Test 2: bilingual CN/EN cross-border listing narration

Request:

{
  "model": "qwen-audio-3.0-tts-plus",
  "input": {
    "text": "欢迎选购 AeroFit 无线耳机。This is our best-selling model on TikTok Shop this month. 电池续航二十八小时，支持主动降噪。Ships worldwide, free returns within 30 days.",
    "voice": "longanlufeng",
    "sample_rate": 44100
  }
}

Real result:

{
  "taskId": "tk-hiapi-01M33PS3GPV0GBXBSHYDMAEA5A",
  "status": "success",
  "model": "qwen-audio-3.0-tts-plus",
  "created": 1790052175,
  "completed": 1790052184,
  "storage": "temp",
  "output": [
    {
      "type": "audio",
      "url": "https://temp.hiapi.ai/7c6ttvrbpt/01M33PS3GPV0GBXBSHYDMAEA5A-0.mp3",
      "artifactId": "187562",
      "expireAt": 1790656984
    }
  ]
}

This script is 138 characters and mixes Chinese and English mid-sentence — there's no separate language flag to set. The model just reads whatever is in text, switching cleanly between the two languages in the same clip. Generation took 9 seconds and cost about $0.0055.

Both clips above have almost certainly expired by the time you're reading this: expireAt is roughly 7 days after created, because the default storage mode is temp. That's expected behavior, not a bug — treat these as a record of real output quality and timing, and in production, download output[0].url and re-host it on your own storage the moment the task completes.

Schema and pricing, briefly

The request schema is small and strict (extra fields are rejected):

• text (string, required) — the script to read
• voice (enum, required) — longanlingxin or longanlufeng, the only two available
• format (enum, optional) — mp3 (default), wav, pcm, or opus
• sample_rate (enum, optional) — one of 8000, 16000, 22050, 24000, 44100, 48000
• pitch (number, optional) — 0.5 to 2
• volume (number, optional) — 0 to 100

There's no emotion, speed, or SSML control on this model — tone and pacing come entirely from how you punctuate and phrase the script.

Pricing is $0.04 per 1,000 characters, confirmed against HiAPI's live pricing endpoint. A single 150-character product blurb costs well under a cent; a catalog of 500 short blurbs at that length works out to about $3 total. The Qwen-Audio 3.0 TTS Plus model page has a live Playground if you want to test your own script and hear both voices before wiring up the API.

Wiring it into a listing or video workflow

1. Write (or localize) the script exactly as it should be read aloud.
2. POST /v1/tasks with the model, your script in text, and a voice.
3. Poll GET /v1/tasks/<id> — or use a callback — until status is success.
4. Download output[0].url immediately; don't store the temp link.
5. Re-host the MP3 on your own storage or CDN.
6. Drop it into your video editor's voice track, or attach it directly to the listing page.

Set an Idempotency-Key header on the create-task request so a retried call after a timeout doesn't generate (and bill) a duplicate clip. For the full field-by-field request and response walkthrough, see our qwen-audio-3.0-tts-plus API tutorial.

What it won't do

• No emotion or speed parameter. If a line needs to land with more energy, rewrite it with shorter sentences and clearer punctuation rather than looking for a tone setting.
• Audio only. It won't touch the visual side of a listing — pair it with an image or video model for photos and product video footage.
• Temporary storage by default. Roughly a 7-day link lifetime unless you request persistent storage; build the re-host step into your pipeline from day one.
• Two voices, no more. If a consistent brand voice across the whole catalog matters, that's actually a constraint you can rely on — the same two voices apply to every SKU, not a random pick each time.

FAQ

Does qwen-audio-3.0-tts-plus support emotional or expressive speech?

No. The schema exposes pitch (0.5–2) and volume (0–100) but no emotion, style, or SSML controls. Expressiveness comes from how the script itself is written and punctuated.

Can qwen-audio-3.0-tts-plus read mixed Chinese and English text in one request?

Yes — our second test above mixed Chinese and English in the same text field and the model switched between them cleanly, with no separate language parameter to set.

How long do qwen-audio-3.0-tts-plus output files stay available?

By default, storage is temp and expireAt lands roughly 7 days after the task completes. Download and re-host the file before that window closes if you need it long-term.

What audio formats does qwen-audio-3.0-tts-plus support?

mp3 (the default), wav, pcm, and opus, set via the optional format field.

Is there a cheaper Qwen-Audio TTS tier?

Yes, a sibling tier, Qwen-Audio 3.0 TTS Flash, is priced at $0.03 per 1,000 characters versus $0.04 for Plus. Both share the same task-based API shape.

Try it yourself

If your listings need narration rather than more images, Qwen-Audio 3.0 TTS Plus is a fast, inexpensive way to get there — run one script through the Playground on the model page to hear it in each voice before scripting a full catalog. For other speech and music models, browse HiAPI's AI audio APIs collection.