Fish Audio Create Voice Model API - Custom TTS Voice

Upload reference audio to create a custom TTS voice model. Synchronous endpoint: the model is ready for speech synthesis immediately after creation.

Upload reference audio to create a custom TTS voice model. On success the response includes model details; use _id as reference_id on the Text to Speech endpoint.

Synchronous endpoint: this request uploads files. For regular REST calls use multipart/form-data and let the HTTP client set the multipart Content-Type boundary automatically. train_mode only supports fast; the model is ready immediately after creation.

Base URL

https://api.autorouter.top — Production

Authentication

BearerAuth: http (bearer) Authenticate using a Bearer Token. Format: Authorization: Bearer sk-xxxxxx

Endpoints

POST `/fishaudio/model`

Create a voice model

Upload reference audio and model metadata; the response returns the new model details.

Because files are uploaded, use multipart/form-data and do not set the boundary manually.

Request Body

Content-Type: multipart/form-data

Example:

{
  "type": "tts",
  "title": "My Voice",
  "description": "Cloned from a studio sample",
  "visibility": "private",
  "train_mode": "fast",
  "voices": ["sample.wav"],
  "texts": ["Hello, this is a studio sample."],
  "tags": ["custom", "english"],
  "enhance_audio_quality": true,
  "generate_sample": false
}

Properties:

Name	Type	Required	Description
`type`	string	Yes	Model type, always `tts` (text to speech)
`title`	string	Yes	Model title or name
`train_mode`	string	Yes	Training mode, always `fast`. TTS models are ready immediately after creation
`voices`	file \| file[]	Yes	Reference audio used to tune the model, 1–20 files
`description`	string \| null	No	Model description
`cover_image`	file \| null	No	Cover image. Required if the model is public
`visibility`	string	No	Visibility. Default `private`
`texts`	string \| string[] \| null	No	Transcripts corresponding to `voices`; if omitted, ASR is run on the audio
`voice_design_signatures`	string \| string[] \| null	No	Signatures from Voice Design candidates, one per voice in the same order
`tags`	string \| string[] \| null	No	Model tags
`enhance_audio_quality`	boolean	No	Whether to enhance audio quality. Default `true`
`generate_sample`	boolean	No	Whether to generate default preview text. Default `false`

visibilitystring(Optional)

Model visibility. Default private.

private: visible only to the creator
unlist: anyone with the link can access
public: shown publicly. API requests with public are downgraded to private; use the web publish flow to publish publicly

Public models require cover_image.

voicesfile | file[](Required)

Reference audio files for tuning. Minimum 1, maximum 20.

Single file: field name voices, upload one audio file
Multiple files: repeat the voices field, one reference clip per file

If texts is omitted, the server runs ASR on the audio. Providing accurate transcripts improves clone quality.

textsstring | string[] | null(Optional)

Transcripts corresponding to voices, up to 20 items. Order must match voices.

If omitted, the server runs ASR on the audio
If provided, use the exact spoken text; accuracy directly affects clone quality

voice_design_signaturesstring | string[] | null(Optional)

Signatures from /v1/voice-design candidates, matching voices in count and order.

When every uploaded voice verifies, the model is stamped source=voice_design; an invalid signature rejects the request.

Responses

201: Created successfully

Content-Type: application/json

401: Unauthorized

Content-Type: application/json

503: Service overloaded

Content-Type: application/json

Request Example

curl -X POST "https://api.autorouter.top/fishaudio/model" \
  -H "Authorization: Bearer sk-xxxxxx" \
  -F "type=tts" \
  -F "title=My Voice" \
  -F "description=Cloned from a studio sample" \
  -F "visibility=private" \
  -F "train_mode=fast" \
  -F "voices=@/path/to/sample.wav" \
  -F "texts=Hello, this is a studio sample." \
  -F "tags=custom" \
  -F "enhance_audio_quality=true"

Response Example

{
  "_id": "7f8a9b0c1d2e3f4a5b6c7d8e",
  "type": "tts",
  "title": "My Voice",
  "description": "Cloned from a studio sample",
  "cover_image": "",
  "train_mode": "fast",
  "state": "trained",
  "tags": ["custom"],
  "samples": [],
  "created_at": "2026-09-20T01:00:00.000Z",
  "updated_at": "2026-09-20T01:00:00.000Z",
  "languages": ["en"],
  "visibility": "private",
  "lock_visibility": false,
  "default_text": "",
  "source": null,
  "licensed": false,
  "like_count": 0,
  "mark_count": 0,
  "shared_count": 0,
  "task_count": 0,
  "liked": false,
  "marked": false,
  "unliked": false,
  "author": {
    "_id": "64a1b2c3d4e5f67890123456",
    "nickname": "creator",
    "avatar": "https://example.com/avatar.png"
  }
}

Response fields:

Name	Type	Description
`_id`	string	Model ID, usable as TTS `reference_id`
`type`	string	Model type. `tts` is text-to-speech; `svc` is singing voice conversion
`title`	string	Model title
`description`	string	Model description. Defaults to an empty string
`cover_image`	string	Cover image URL. Defaults to empty
`train_mode`	string	Training mode. `fast` is ready immediately after creation; `full` is full training
`state`	string	State. Enum: `created`, `training`, `trained`, `failed`
`tags`	string[]	Tag list
`samples`	object[]	Preview samples. Defaults to an empty array
`created_at`	string	Created at, ISO 8601
`updated_at`	string	Updated at, ISO 8601
`languages`	string[]	Language list
`visibility`	string	Visibility. Enum: `public`, `unlist`, `private`
`lock_visibility`	boolean	Whether visibility is locked. Default `false`
`dmca_taken_down`	boolean \| null	Whether the model was taken down for DMCA
`takedown_category`	string \| null	Takedown category: `dmca`, `policy`
`default_text`	string	Default preview text
`source`	string \| null	Source. May be `voice_design` when every uploaded voice verifies
`licensed`	boolean	Whether this is a Fish Audio licensed voice
`quality`	object \| null	Audio quality evaluation
`like_count`	integer	Like count
`mark_count`	integer	Mark count
`shared_count`	integer	Share count
`task_count`	integer	Usage count
`liked`	boolean	Whether the current user liked it
`marked`	boolean	Whether the current user marked it
`unliked`	boolean	Whether the current user unliked it
`author`	object	Author information

authorobject(Required)

Author information.

_idstring(Required)

Author user ID.

nicknamestring(Required)

Author nickname.

avatarstring(Required)

Author avatar URL.

samplesobject[](Optional)

List of preview samples. Defaults to an empty array.

titlestring(Required)

Sample title.

textstring(Required)

Text corresponding to the sample.

task_idstring(Required)

Task ID that generated this sample.

audiostring(Required)

Sample audio URL.

Error Handling

HTTP 401 / 403 Auth Errors

401 Unauthorized: API Key is invalid or expired
403 Forbidden: API Key cannot access this model (check the token model allowlist)

HTTP 402 Insufficient Balance

Returns insufficient user quota. Top up in the AutoRouter console.

HTTP 503 Service Overloaded

The server is too busy to process the request. Please retry later.

Create Model

Base URL

Authentication

Endpoints

POST `/fishaudio/model`

Request Body

visibilitystring(Optional)

voicesfile | file[](Required)

textsstring | string[] | null(Optional)

voice_design_signaturesstring | string[] | null(Optional)

Responses

Request Example

Response Example

authorobject(Required)

samplesobject[](Optional)

Error Handling

HTTP 401 / 403 Auth Errors

HTTP 402 Insufficient Balance

HTTP 503 Service Overloaded

On this page