Skip to main content
Use this page as the source of truth for publicly documented TTS model IDs.

Model overview

Public model ID list

Model V3.5 High Quality (myvocal_v3_5)

Use this public alias with either existing TTS endpoint. Existing defaults, character metering, plans and authentication stay unchanged. Use an existing voice you are permitted to use; the generic voices endpoint does not promise a model-specific compatibility filter for this model.
  • Omit language (or send auto) for automatic detection. Explicit codes use the model-specific table below. Declared language support is not a claim that every voice/language combination has been tested.
  • Default voiceSettings.stability is 0.5; the existing inclusive legal range 0.1–1 remains accepted, including explicit 0.1, 0.5 and 1. Other models retain their defaults.
  • Text tags such as [whispering], [shouting] and [laughing] can guide delivery; their effect varies. Keep spoken text with tags. This adds no SSML, speed, style or similarity API field.
  • Nonstream responses contain the complete audio. A stream may terminate if generation or delivery fails after audio begins; do not treat a truncated stream as a completed recording or expect JSON appended to its audio bytes.
  • Before audio is sent, generation errors use the existing MyVocal error envelope (10003), invalid languages 10004, unavailable voices 10005, and invalid stability 10006.
  • Only the standard High Quality product is included. No new default model or price multiplier is introduced.

Model V3.5 language codes

Open Model V3.5 language codes

View the independent list of 87 explicit language codes and Auto behavior.

V3 language scope

  • Supported language count: 98
  • Excluded from support: Kirghiz, Lingala

Open V3 language code list

View the complete supported language and code table.

Accent model (myvocal_v3_accent_enhance)

The Accent model has its own language contract. It is not the same set as myvocal_v3.
  • Supported language count: 40 language codes, plus auto.
  • Omitting language is equivalent to auto.
  • A code outside this list is rejected with code = 10004. This includes plain V3 codes the Accent model does not support, such as hmn.
  • Legacy mapped enum values (for example US_ENGLISH) are not Accent codes. Use en.
Accepted Accent language codes:
Other Accent behavior:
  • Voices: use the compatible voices returned by GET /sound_clone/api/v1/voices?modelId=myvocal_v3_accent_enhance. That list includes your own compatible voices and the prebuilt voices available to your API key. A voice listed there is a selection aid, not a guarantee that a later generation will succeed.
  • voiceSettings.stability: a value inside the legal range (0.1 to 1) is accepted for compatibility but has no effect on this model. Out-of-range values are still rejected with code = 10006.
  • Prompt tags: inline [tag] prompt control is documented for myvocal_v3. It is not documented as equivalent for the Accent model; send plain text.
  • Character metering, plan limits and rates stay the same as the other TTS models.

Default path vs explicit modelId

  • Default path: omit modelId and follow current V2/default mapping rules.
  • Explicit V3 path: set modelId to myvocal_v3.
  • Explicit Accent path: set modelId to myvocal_v3_accent_enhance.
English(US) currently uses the default model path. A public explicit modelId for English(US) is not published in this phase.

V2 to V3 migration guidance

  1. Keep your V2 flow as fallback.
  2. Add a feature flag for modelId: "myvocal_v3".
  3. Use explicit language if you want deterministic target language output.
  4. Omit language for auto-detect behavior on V3.
  5. Roll out with traffic slices and monitor output quality.

Migration to the Accent model

The table describes what changes when you move from the default/V2 path or from myvocal_v3 to myvocal_v3_accent_enhance.