Voice Cloning API: Clone a Voice in One HTTP Call

Create an instant voice clone via a single POST call. Uploads the audio, runs preprocessing, and returns pre-generated sample clips of the cloned voice - all in one request.

Requirements

• A Smallest AI API key - grab one from the API Keys page.
• A clean audio sample, 5–15 seconds, under 5 MB. Supported types: .mp3, .wav, .mp4, .webm.

export SMALLEST_API_KEY=sk_your_key_here

Create the clone

curl -X POST "https://api.smallest.ai/waves/v1/voice-cloning" \
  -H "Authorization: Bearer $SMALLEST_API_KEY" \
  -F "displayName=my-custom-voice" \
  -F "file=@/path/to/sample.wav" \
  -F "model=lightning-v3.1-pro" \
  -F "language=en" \
  -F "accent=general" \
  -F "tags=english,friendly"

Request fields

Response shape

{
  "message": "Voice clone created successfully",
  "data": {
    "voiceId": "voice_dLP5T67Qw7",
    "displayName": "my-custom-voice",
    "status": "completed",
    "language": "en",
    "model": { "_id": "...", "modelName": "lightning-v3.1-pro" },
    "audioFileNames": ["sample.wav"],
    "createdAt": "2026-04-20T09:42:17.103Z",
    "organizationId": "69561896a37fd214b9a8d33c",
    "samples": [
      {
        "text": "Hello! This is a sample of what I sound like.",
        "audioUrl": "https://..."
      }
    ]
  }
}

Use data.voiceId directly in any TTS call:

curl -X POST "https://api.smallest.ai/waves/v1/tts" \
  -H "Authorization: Bearer $SMALLEST_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Hello, this is my cloned voice.",
    "voice_id": "voice_dLP5T67Qw7",
    "model": "lightning_v3.1_pro"
  }' \
  -o output.wav

List your clones

curl -H "Authorization: Bearer $SMALLEST_API_KEY" \
  "https://api.smallest.ai/waves/v1/voice-cloning"

Returns every clone on your organization along with modelIds (compatible models), status, language, and creation time.

Check model compatibility before using a clone

The modelIds array in the list response tells you which TTS models a clone works with. Check it before passing voice_id to a TTS call:

curl -H "Authorization: Bearer $SMALLEST_API_KEY" \
  "https://api.smallest.ai/waves/v1/voice-cloning" \
  | jq '.data[] | {displayName, voiceId, modelIds}'

{ "displayName": "my-voice",   "voiceId": "voice_abc", "modelIds": ["lightning-v3.1-pro"] }
{ "displayName": "old-voice",  "voiceId": "voice_xyz", "modelIds": ["lightning-large"] }

Clones created via the current POST /waves/v1/voice-cloning endpoint produce voices for the pool you pass in model. Pair the resulting voice_id with the matching TTS model; check modelIds on the clone if unsure. The older POST /waves/v1/lightning-large/add_voice endpoint (deprecated) only produces lightning-large-compatible voices, which will not work with the current pools.

Delete a clone

The current public delete endpoint lives at /waves/v1/lightning-large despite the path suggesting otherwise - it deletes any voice clone on your organization regardless of the underlying model.

curl -X DELETE "https://api.smallest.ai/waves/v1/lightning-large" \
  -H "Authorization: Bearer $SMALLEST_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "voiceId": "voice_dLP5T67Qw7" }'

Errors

Migrating from the legacy endpoint

If you’re currently using POST /waves/v1/lightning-large/add_voice, switch to POST /waves/v1/voice-cloning:

• Same auth - Authorization: Bearer $SMALLEST_API_KEY.
• Same multipart shape for displayName + file.
• New optional fields - language, description, accent, tags.
• Response includes samples - pre-generated audio clips of the cloned voice, which the legacy endpoint did not provide.
• Pass model=lightning-v3.1-pro. Voices cloned via the legacy endpoint only worked on lightning-large and returned empty audio on the current pools.

The legacy endpoint still works but is marked deprecated in the API reference.

Need help? Email support@smallest.ai or ask on Discord.

FAQ: voice cloning API

How much audio do I need to clone a voice? A clean 5 to 15 second sample under 5 MB, as .mp3, .wav, .mp4, or .webm. Recording quality matters more than length; see recording guidelines.

How do I use the cloned voice in text to speech? Pass the returned voiceId as voice_id in any TTS call, with the TTS model that matches the pool you cloned onto. The example above shows the full round trip.

Can I clone onto the premium voice pool? Yes. Pass model=lightning-v3.1-pro when creating the clone, then use the voice with the lightning_v3.1_pro TTS model.

Why does my cloned voice return a tiny or empty audio file? Almost always a lightning-large-only clone used on the current pool. Check the clone’s modelIds in the list response and recreate it on lightning-v3.1-pro if needed.