Docs / Guides
TTS Workbench Guide
The homepage chat box suits "paste something and read it on a whim." The TTS Workbench (`/studio`) suits "I want to seriously pick a voice" — it lays out all voices from all eight brands as a filterable, paginated voice library.
The homepage chat box suits "paste something and read it on a whim." The TTS Workbench (
/studio) suits "I want to seriously pick a voice" — it lays out all voices from all eight brands as a filterable, paginated voice library.
How to Get In
Enter TTS Workbench from the homepage's bottom-left menu, or go to /studio directly.
At the top of the workbench are two mode switches: TTS (speech synthesis) and Translation (see the translation guide).
Interface Layout
| Area | Purpose |
|---|---|
| Top toolbar | Search voices, filter by group/sex/language |
| Middle voice-card area | The voice library itself; each card is one voice; click a card to preview |
| Right settings panel | Choose brand, model, preview text, speech rate, consumption-rate display |
| Bottom pagination | Brands with many voices are paginated; adjust items per page |
Picking a Voice in Four Steps
1. Pick the Brand and Model First
At the top of the right settings panel, choose a brand (the eight + system voice), then a model tier.
Immediately read the "~N credits / 1,000 chars" in the settings panel — that's the real consumption rate for this tier. Switching models within the same brand can change the price several-fold (e.g. Doubao 2.0 is 48, 1.0 is 80).
Brands with a small dot marker on the brand selector have a free-preview voice.
2. Narrow Down With the Filters
Brands with many voices (like Doubao, Microsoft) have hundreds at once — paging through directly is inefficient. Use the top toolbar:
- Search box: search by voice name
- Group: e.g. "Super-Natural LLM," "Premium Voice Library," "System Voices"
- Sex: male / female / child
- Language: Chinese / English / Japanese…
3. Change the Preview Text
The preview text box in the right settings panel defaults to a sample sentence, but you can change it to whatever you actually want to read — previewing with your real text is far more accurate than listening to the official sample.
Exception: if you haven't subscribed and are using a voice from the free preview whitelist, that script is fixed and can't be changed (a necessary limit to prevent free-quota abuse). After subscribing you can preview with any text.
4. Preview and Confirm
Click a voice card to play it. Once you've picked one, the voice is remembered and used when you read on the homepage.
About the Speech Rate
The speech rate is adjusted in the right settings panel, in 5 levels: 0.75x / 1x / 1.25x / 1.5x / 2x.
One thing to know: for cloud AI voices, the rate is "baked into" the audio during server-side synthesis — it can't be changed on the fly like system voices during playback. So after changing the rate, it needs to be re-synthesized to hear the effect (Soniva handles this automatically; you'll just feel a brief reload).
Each upstream vendor's rate parameter works differently (some use a 0.5–2 multiplier, some an integer 0–9 scale, and Cartesia even uses a relative -1~1 value). Soniva normalizes them all into these 5 levels. So the same rate level can differ slightly in actual speed across brands — that's decided by the upstream interface.
Advanced Settings (Usually Unneeded)
At the bottom of the right settings panel is an "Advanced settings" area for users with special needs:
- Qwen instruction: Alibaba Cloud
qwen3-tts-instruct-flashsupports describing the desired tone in natural language - Timeout: lengthen it when the upstream responds slowly
- Workspace ID / Region / WebSocket endpoint: advanced params for Alibaba Cloud Bailian workspace keys
- Custom request JSON: directly control what's sent to upstream
If you don't know what these are, leave them alone — leaving them blank, Soniva uses the defaults.
System Voices Are Here Too
Besides the eight cloud brands, the brand list also includes your device's built-in system voices (Apple / Google / Microsoft, depending on your device and browser).
System voices are completely free, unlimited, and don't require login — you can preview and pick them in the workbench too. See voice engines.
Related Documents
Was this article helpful?
Anonymous feedback is enabled when you allow analytics and the service is available. You can also report an issue by email.
The email includes this article. You can also copy the details and send them to soniva.cde@gmail.com