Audio

# speech synthesis
curl https://api.ailoot.ai/v1/audio/speech \
  -H "Authorization: Bearer $AILOOT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "qwen-audio-3.0-tts-plus", "input": "Good afternoon! This is a text-to-speech example.", "voice": "longanlingxin"}' \
  --output speech.mp3

# transcription
curl https://api.ailoot.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $AILOOT_API_KEY" \
  -F model=qwen-asr \
  -F file=@meeting.wav

# speech transcription
with open("meeting.wav", "rb") as f:
    tr = client.audio.transcriptions.create(model="qwen-asr", file=f)
print(tr.text)

{
  "text": "Good afternoon, everyone. Let us start with the results for the week…"
}

Authentication

• Authorizationheaderrequired

  The key in the format Bearer sk-…. The following headers also work: x-api-key or x-goog-api-key.

Text-to-speech: request body

• modelstringrequired

  Text-to-speech model: gemini-3.8-flash-tts, qwen-audio-3.0-tts-plus, speech-2.8-turbo, gemini-3.8-flash-lite-tts, speech-2.8-hd, gemini-3.1-flash-tts-preview, speech-2.6-turbo, speech-2.6-hd, speech-02-turbo, speech-02-hd.

• inputstringrequired

  Text to convert to speech. MiniMax models accept fewer than 10,000 characters; qwen-audio-3.0-tts-plus accepts up to 600.

• voicestring

  Voice. Required for MiniMax models (speech-2.8-hd and other models matching speech-…): a MiniMax voice ID (Latin letters, digits, _ and -, up to 128 characters). For qwen-audio-3.0-tts-plus the default is longanlingxin; for Gemini TTS models, it is Kore.

• response_formatstringoptional

  MiniMax supports only mp3; qwen-audio-3.0-tts-plus — mp3, wav or pcm; Gemini TTS supports wav or json (a link to the file).

The response is an audio file (for MiniMax, audio/mpeg). Billing is per text character or per token, as listed in the catalog.

Speech-to-text

• filefilerequired

  A WAV audio file in multipart/form-data. qwen-asr, qwen-asr-latest and qwen-audio-3.0-asr-flash: a complete, uncompressed PCM WAV file up to 5 minutes long and approximately 7.5 MB (up to 10 MB in base64). asr-1.0: PCM WAV up to 500 seconds and 50 MB. gemini-3.5-transcribe: WAV up to 64 MB.

• modelstringrequired

  Speech-to-text model: gemini-3.5-transcribe, qwen-audio-3.0-asr-flash, asr-1.0.

• response_formatstringoptional

  Only json (or omit it).

The field language is not used by these models. Response: {"text": "…"}. Billing is based on recording duration (seconds, minutes, or hours — each model uses its own unit).

Realtime voice

For qwen-audio-3.0-realtime-plus open a WebSocket wss://api.ailoot.ai/v1/realtime?model=qwen-audio-3.0-realtime-plus with the header Authorization: Bearer <API_KEY>. Send the key from your server: keys in the query string are not accepted. Wait for session.created with your chosen model, then send the settings, a message, and response.create. Below are three consecutive JSON text frames for a text response in an audio session:

{"type":"session.update","session":{"modalities":["text"],"turn_detection":null,"max_history_turns":20,"instructions":"Be brief."}}

{"type":"conversation.item.create","item":{"type":"message","role":"user","content":[{"type":"input_text","text":"Say yes."}]}}

{"type":"response.create"}

Read response events until you receive response.done and its final usage; errors arrive in an event of type error. Billing is based on the text and audio tokens in each confirmed response. Audio input: send an event of type input_audio_buffer.append with audio — base64 PCM16 mono 16 kHz; then input_audio_buffer.commit and response.create. Audio output is PCM16 mono 24 kHz. Binary frames are not accepted in this session.

Supported client events: session.update, input_audio_buffer.append/commit/clear, response.create/cancel, conversation.item.create/retrieve/delete. Conversation items support only messages/text; function tools, web search, and voiceprint are not supported. Session limits: 600 seconds, 20 responses, 20 history turns, 2000 bytes of instructions, and 8000 bytes of client text. The idle timeout depends on the configuration (60 seconds by default).