Audio Models

Atlas Cloud exposes three different audio capabilities — text-to-speech, music generation, and speech-to-text — through a single endpoint. Which one you get is determined by the model you choose, not by the URL.

POST https://api.atlascloud.ai/api/v1/model/generateAudio

How it works

Audio requests are asynchronous, like image and video generation:

See Predictions for the full status machine and Webhooks for callback setup.

Text-to-speech

Synthesize speech from text. Available models include Seed Audio, ElevenLabs, MiniMax Speech, Gemini TTS, and xAI TTS.

Example: Seed Audio 1.0

curl -X POST https://api.atlascloud.ai/api/v1/model/generateAudio \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance/seed-audio-1.0",
    "text": "Welcome to Atlas Cloud.",
    "format": "mp3",
    "sample_rate": 24000
  }'

To clone or reference a voice, attach a reference clip and point at it from the text:

{
  "model": "bytedance/seed-audio-1.0",
  "text": "Use the voice of @audio1 and say: your order has shipped.",
  "references": [{ "audio_url": "https://example.com/sample-voice.mp3" }]
}

Reference clips must be 30 seconds or less and under 10 MB.

Music generation

Compose full tracks, with or without vocals. Suno Chirp and MiniMax Music are available, plus a dedicated lyric-writing model.

Example: Suno Chirp v5

curl -X POST https://api.atlascloud.ai/api/v1/model/generateAudio \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "suno/chirp-v5",
    "prompt": "an upbeat indie pop song about summer road trips",
    "vocal_gender": "Female"
  }'

To write lyrics first and compose second, call a lyric model such as minimax/lyrics-generation, then pass its output into the music model's lyrics field.

Speech-to-text

Transcribe recordings, with optional speaker separation and word-level timestamps.

Example: Seed ASR 2.0

curl -X POST https://api.atlascloud.ai/api/v1/model/generateAudio \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance/seed-asr-2.0",
    "audio_url": "https://example.com/interview.mp3",
    "enable_punc": true,
    "enable_speaker_info": true,
    "show_utterances": true
  }'

Example: xAI STT v1

curl -X POST https://api.atlascloud.ai/api/v1/model/generateAudio \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "xai/stt-v1",
    "audio": "https://example.com/interview.mp3",
    "diarize": true,
    "text_normalization": true
  }'

Reading the result

{
  "code": 200,
  "data": {
    "id": "…",
    "status": "completed",
    "outputs": ["Thanks for joining the call today…"],
    "stt_result": {
      "text": "Thanks for joining the call today…",
      "duration": 184.2,
      "words": [
        { "text": "Thanks", "start": 0.12, "end": 0.41, "speaker_id": "1" }
      ]
    }
  }
}

outputs[0] holds the plain transcript. Structured output — timings, speakers, detected language — lives in stt_result.

Providing audio input

Audio input fields accept either a public HTTPS URL or a Base64 data URI. Base64 inputs are stored automatically and replaced with a URL before the request reaches the model.

For local files, upload them first and use the returned URL:

curl -X POST https://api.atlascloud.ai/api/v1/model/uploadMedia \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -F "[email protected]"

See Upload Files for limits and details.

Billing

Audio models are billed in one of three ways, depending on the model:

Failed jobs are not billed. For models billed by output duration, an amount is held when the job is submitted and settled against the real duration on completion.

Use the pricing endpoint to get an exact quote before generating, and see Model Billing for worked examples.

Finding audio models

The model catalog changes frequently. Rather than hard-coding a list, filter the catalog by type:

• Browse the Model Library and filter by Text-to-Audio or Audio-to-Text
• Or list them from an agent with the MCP Server using atlas_list_models with type="Audio"

Every model's exact parameters are published in its own API reference page under Model endpoints.

Related