Design a voice without a reference

Before you start

• For local design, use a matching Qwen3-TTS VoiceDesign package. Base and CustomVoice do not replace VoiceDesign.
• Alternatively, configure a MiniMax cloud account with permission to create voices. Cloud and local forms are configured separately.
• A voice description and a short audition passage. This workflow does not need a reference recording.

Design and read with Qwen cloud audio

Qwen Audio 3.1 can generate a voice from a description without a reference recording or local voice model. Your account needs model access. Generate sends the text and description to the service, which may charge for the request. The result is a complete recording; listen to judge its voice and expression.

1. 01

   Save a usable cloud connection

   Open the API services step in Setup Wizard, search for Qwen Audio 3.1 and enable it. Enter https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com, replacing the bracketed part with your actual WorkspaceId. Do not append an API path, port or /v1. Enter your API key and qwen-audio-3.1-tts-next in the speech synthesis model field, then save. Keep your existing recognition model in its separate field. No local model download is needed here.

   123

   1. 1Enter your actual workspace root, without an API path.
   2. 2Enter your own API key; the screenshot uses a local example.
   3. 3Enter qwen-audio-3.1-tts-next in TTS; the ASR field remains separate.

2. 02

   Choose Voice design and fill both fields

   Open Voice Tools → Model Setup → Cloud API, choose Qwen Audio 3.1, select your Saved connection, then choose Voice design. Under Voice and emotion description, describe the voice, delivery and rhythm you want, such as a warm, natural voice with a relaxed pace and a gentle smile. Under Text to read, enter the actual spoken words. The combined prompt is limited to 3,000 characters. Leave out reference markers, tabs and other control characters.

   12345

   1. 1Choose the saved Qwen TTS connection.
   2. 2Choose Voice design; no reference recording is needed.
   3. 3Describe the voice, expression and pace separately from the spoken words.
   4. 4Enter the spoken text; the combined prompt is limited to 3,000 characters.
   5. 5Generate and listen to judge the description and pronunciation.

3. 03

   Try a short passage and adjust

   Select Generate voice and listen when the player appears. Edit the description and generate again if needed. Download WAV when satisfied, or keep takes in A/B for comparison. Expand Rate and volume when you need those controls. After Cancel, wait for settlement before editing. This feature generates an audition recording, does not create an account voice ID and does not feed live interpretation directly.

4. 04

   Use Read aloud for plain text

   Select Read aloud, enter your words and select Generate voice. This mode sends no voice description or reference recording. Listen and Download WAV as usual. To use a person's reference recording, follow the reference cloning guide.

   1234

   1. 1Choose a connection saved in setup; do not enter its address or key again.
   2. 2Choose Read aloud to send only the text.
   3. 3Enter the words you want spoken.
   4. 4Generate calls the cloud service; listen and download WAV after completion.

Use local models and other services

1. 01

   Choose voice design and its model

   Save the matching Qwen3-TTS VoiceDesign folder or GGUF, size and quantization in Setup Wizard → Model library. Open Voice Tools → Model Setup → Local models, choose the saved model and select Describe a voice · VoiceDesign. Select the saved configuration here rather than entering its path again.

   12

   1. 1VoiceDesign needs both a description and text.
   2. 2Generate a designed voice, then audition.

2. 02

   Describe the voice, then enter spoken text

   Under Voice and expression description, specify age impression, timbre, pace and delivery, for example: a warm, calm Mandarin female voice with clear articulation and a relaxed pace. Enter the words to read under Synthesis text. Keep description and spoken text separate.

   12

   1. 1VoiceDesign needs both a description and text.
   2. 2Generate a designed voice, then audition.

3. 03

   Check, generate and audition

   Select Check model in the action area above the editor and follow Next step for missing files or components. With checks passed and description/text entered, select Generate designed voice in the same area; no reference recording is needed. Listen after completion and use Cancel task there while running. Generate again after changing the description. File inspection is neither an audition nor inference warm-up; design does not require the optional clone preloading action.

   12

   1. 1VoiceDesign needs both a description and text.
   2. 2Generate a designed voice, then audition.

4. 04

   Save a take, or use cloud design

   Click Save audio when satisfied, or name the take and save it to the voice library. For cloud design, open the design tab under Cloud API → MiniMax. Enter the description and audition text, create the voice, listen and check its returned voice ID before applying it to speech settings.

   123

   1. 1Describe the voice you want to hear.
   2. 2Enter the audition text.
   3. 3Create the voice after checking the description.

If something goes wrong

Next steps