VoiceLab API

Generate text and speech, transcribe or clean up audio, clone your voice, and stream realtime audio.

Keep your API key on your server. Product endpoints use API keys; account management uses a signed-in user's JWT.

Latest STT update

October 9, 2026: STT now accepts 33 languages, including Chinese and Azerbaijani, plus language=auto for automatic detection in uploaded recordings and realtime requests. See the language codes, timing behavior, and availability notes. Existing API routes and request formats are unchanged.

Base URL

text

https://api.voicelab.uz

The public developer API uses the /v1 prefix. For example:

text

POST https://api.voicelab.uz/v1/tts

Choose the right credential

Credential	Used for	Example
User access JWT	VoiceLab website and API-key management	`Authorization: Bearer eyJ...`
Developer API key	Your application calling LLM, TTS, STT, Voice Isolator, Meeting AI, voices, and realtime ticket endpoints	`Authorization: Bearer vlk_...`
Realtime ticket	One short-lived WebSocket connection	`?ticket=eyJ...`

Do not send a user JWT to /v1/tts, /v1/stt, or /v1/voices. Do not send a long-lived API key in a WebSocket URL. The only value allowed in that URL is a short-lived ticket returned by POST /v1/ticket.

Voice Isolator developer requests use an API key at /v1/voice-isolations. The dashboard routes at /api/v1/voice-isolations require a user JWT.

Meeting AI developer requests use a personal workspace API key at /v1/meetings. The dashboard routes at /api/v1/meetings require a user JWT.

Endpoints

Create and manage keys with the signed-in dashboard:

Method	Endpoint	Credential
`GET`	`/api/v1/account/api-key-permissions`	User JWT
`GET`	`/api/v1/account/api-key-scopes`	User JWT
`POST`	`/api/v1/account/api-keys`	User JWT
`GET`	`/api/v1/account/api-keys`	User JWT
`PATCH`	`/api/v1/account/api-keys/{key_id}`	User JWT
`POST`	`/api/v1/account/api-keys/{key_id}/revoke`	User JWT
`DELETE`	`/api/v1/account/api-keys/{key_id}`	User JWT

Use the generated key for application services:

Feature	Endpoints	Permission
LLM models	`GET /v1/models`	`llm:read`
Chat completion	`POST /v1/chat/completions`	`llm:write`
LLM request status	`GET /v1/llm/requests/{id}`	`llm:read`
LLM files	`POST`, `GET /v1/files`, `GET`, `DELETE /v1/files/{file_id}`	`llm:read` for reads, `llm:write` for upload/delete
TTS capabilities	`GET /v1/tts/languages`	TTS read
Voices	`GET /v1/voices?language={code}`	Voices read
Clone discovery/state	`GET /v1/voice-enrollments/capabilities`, `GET /v1/voice-enrollments/{id}`	`voices:read`
Verified voice enrollment	`POST /v1/voice-enrollments`, then `/{id}/transcript`, `/{id}/challenge`, `/{id}/verify`	`voices:write`
Clone metadata	`GET /v1/voices/{id}`	`voices:read`
Edit/delete own clone	`PATCH /v1/voices/{id}`, `DELETE /v1/voices/{id}`	`voices:write`
TTS generation	`POST /v1/tts`	TTS write
Background TTS	`POST /v1/tts/async`	TTS write; TTS read for polling
TTS history	`GET /v1/tts/generations`, `GET /v1/tts/generations/{id}`	TTS read
Delete TTS audio	`DELETE /v1/tts/generations/{id}`	TTS write
STT transcription	`POST /v1/stt`	STT write
STT credit preflight	`GET /v1/stt/preflight`	STT write
STT history and files	`GET /v1/stt/transcriptions...`	STT read
Edit or delete STT	`PATCH`, `PUT`, `DELETE /v1/stt/transcriptions/{id}`	STT write
Voice isolation	`POST /v1/voice-isolations`, `POST /v1/voice-isolations/batch`	`audio_isolation:write`
Isolation history and results	`GET /v1/voice-isolations`, `GET /v1/voice-isolations/{id}`	`audio_isolation:read`
Hide isolation	`DELETE /v1/voice-isolations/{id}`	`audio_isolation:write`
Create isolation export	`POST /v1/voice-isolations/{id}/exports`	`audio_isolation:write`
Get isolation export	`GET /v1/voice-isolations/{id}/exports/{format}`	`audio_isolation:read`
Meeting AI capabilities	`GET /v1/meetings/capabilities`	`meetings:read`
Create meeting	`POST /v1/meetings`	`meetings:write`
Meeting history and results	`GET /v1/meetings`, `GET /v1/meetings/{id}`	`meetings:read`
Rename or delete meeting	`PATCH`, `DELETE /v1/meetings/{id}`	`meetings:write`
Regenerate meeting notes	`POST /v1/meetings/{id}/notes/generations`	`meetings:write`
Realtime TTS	`POST /v1/ticket`, then `GET /v1/tts/stream`	TTS realtime
Realtime STT	`POST /v1/ticket`, then `GET /v1/stt/stream`	STT realtime
Webhooks and deliveries	`/v1/webhooks...`; see complete reference	`webhooks:write`, including reads

Grant Voice Isolator → Access for the complete upload, polling, and export workflow. This enables both isolation scopes. Existing restricted keys need this permission added explicitly. See Voice Isolator.

Grant Meeting AI → Access to upload recordings, poll results, and manage meetings. This enables both meeting scopes. Only personal workspace keys can use Meeting AI. See Meeting AI.

Grant llm: access for model discovery, completions, and request status. Developer LLM requests use API keys and token-based credits. Ask's /api/v1/llm/* and /api/v1/chats/* routes remain JWT-only and uncharged. See LLM API.

Grant Voices → Access for the verified Voice Cloning workflow and owner-only editing/deletion. All active paid plans qualify; consent, a fresh live recording and both verification checks are required.

Analytics and request logs are dashboard views. They use a user JWT and show only traffic authenticated with developer API keys; website JWT traffic is not included. See Analytics and logs.

Quick start

1. Create an API key

Use a user access JWT from the VoiceLab dashboard to create a key. The secret is returned only in this response.

bash

curl -X POST 'https://api.voicelab.uz/api/v1/account/api-keys' \
  -H 'Authorization: Bearer USER_ACCESS_JWT' \
  -H 'Content-Type: application/json' \
  -d '{
    "name": "My production integration",
    "permissions": {
      "text_to_speech": "access",
      "speech_to_text": "access",
      "voices": "read"
    },
    "expires_at": null,
    "restrict_key": true,
    "auto_disable_if_leaked": true
  }'

Save data.secret immediately. It cannot be recovered later.

2. Store the key on your server

bash

export VOICELAB_API_KEY='vlk_replace_with_the_secret_from_creation'

For production, use your deployment secret manager instead of a shell file.

3. Generate speech

bash

tts_http_status=$(curl -fS -w '%{http_code}' 'https://api.voicelab.uz/v1/tts' \
  -H "Authorization: Bearer $VOICELAB_API_KEY" \
  -H 'Content-Type: application/json' \
  -H 'Idempotency-Key: hello-world-001' \
  -d '{
    "text": "Hello from VoiceLab.",
    "language": "en",
    "voice_id": "voice_01J9NEUTRAL0000000000000001",
    "speed": 1
  }' \
  -o tts-response)
case "$tts_http_status" in
  200) mv tts-response hello.wav ;;
  202) cat tts-response ;; # Poll the returned status_url before downloading audio
  *) cat tts-response >&2; exit 1 ;;
esac

Output is mono 24 kHz WAV. Fetch available models and languages from /v1/tts/languages. If generation is pending, the response is 202 JSON instead of audio. See TTS polling.

4. Transcribe audio

bash

curl -fS 'https://api.voicelab.uz/v1/stt' \
  -H "Authorization: Bearer $VOICELAB_API_KEY" \
  -H 'Idempotency-Key: 550e8400-e29b-41d4-a716-446655440000' \
  -F 'audio=@meeting.wav;type=audio/wav' \
  -F 'language=en' \
  -F 'include_speakers=false'

See Speech to Text for upload limits and polling.

Common response rules

JSON responses use Content-Type: application/json.
Resource timestamps are RFC 3339 UTC strings, for example 2026-08-16T12:30:00Z. LLM completion created and file created_at/expires_at use Unix seconds; model created is 0.
IDs, cursors, tickets, and URLs are opaque. Do not parse or construct them.
Standard resource responses include request_id. LLM success bodies use their own shapes; completion headers include X-LLM-Request-ID. LLM streaming uses SSE. See LLM response formats.
TTS returns WAV audio or pending-job JSON. STT exports return files. Voice Isolator returns JSON with signed download URLs.
Private audio URLs are short-lived signed URLs. Do not store them as permanent links.

Error response

Standard JSON errors use this shape:

json

{
  "message": "This API key is not allowed to perform this operation.",
  "error": {
    "code": "insufficient_scope"
  },
  "request_id": "req_01J..."
}

LLM handler errors put the message in error.message and include error.type. Shared authentication errors can still use the standard shape shown above. Handle both envelopes when integrating LLM endpoints.

Some validation errors also contain error.fields:

json

{
  "message": "Check the highlighted fields.",
  "error": {
    "code": "validation_error",
    "fields": {
      "language": "Choose a supported language."
    }
  },
  "request_id": "req_01J..."
}

Always log request_id in your application. Never log an API key, realtime ticket, audio bytes, or signed audio URL.

Voice Cloning

VoiceLab API

Latest STT update

Base URL

Choose the right credential

Endpoints

Quick start

1. Create an API key

2. Store the key on your server

3. Generate speech

4. Transcribe audio

Common response rules

Error response

Documents

Authentication and API keys

LLM

LLM files

Voices and capabilities

Voice Cloning

Webhooks

Text to Speech

Speech to Text

Voice Isolator

Meeting AI

Realtime TTS

Realtime STT

Analytics and logs

Errors and limits

VoiceLab API ​

Latest STT update ​

Base URL ​

Choose the right credential ​

Endpoints ​

Quick start ​

1. Create an API key ​

2. Store the key on your server ​

3. Generate speech ​

4. Transcribe audio ​

Common response rules ​

Error response ​

Documents ​

VoiceLab API

Latest STT update

Base URL

Choose the right credential

Endpoints

Quick start

1. Create an API key

2. Store the key on your server

3. Generate speech

4. Transcribe audio

Common response rules

Error response

Documents