Speech

POST /v1/audio/speech reads text aloud and returns the audio in the response body, like OpenAI’s speech endpoint. It is synchronous: there is no job to poll.

curl https://api.geekidea.ai/v1/audio/speech \
  -H "Authorization: Bearer $GEEKIDEA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen3-tts-flash",
    "input": "The weather is lovely today. Let'\''s take a walk in the park.",
    "voice": "Cherry",
    "response_format": "wav"
  }' \
  --output speech.wav

The response is the audio file (Content-Type: audio/wav) with these headers:

Header	Description
`x-geekidea-id`	The call ID (`audio_…`), also its billing ID
`x-geekidea-characters`	Billed characters
`x-geekidea-cost`	Amount charged, in USD

With the OpenAI SDK:

import os

from openai import OpenAI

client = OpenAI(base_url="https://api.geekidea.ai/v1", api_key=os.environ["GEEKIDEA_API_KEY"])
with client.audio.speech.with_streaming_response.create(
    model="qwen/qwen3-tts-flash", voice="Cherry", input="Hello from GEEKIDEA.", response_format="wav"
) as response:
    response.stream_to_file("speech.wav")

Parameters

Field	Description
`model`	Required. A model whose `type` is `audio`
`input`	Required. The text to read, up to the model’s `generation_config.max_input_characters` (at most 10,000)
`voice`	A value from the model’s `generation_config.voice`. Defaults to the model’s default
`response_format`	A value from the model’s `generation_config.response_format`. Currently `wav`
`language`	GEEKIDEA extension. A value from the model’s `generation_config.language`, such as `Chinese` or `English`. Detected automatically when omitted
`metadata`	Up to 16 string key–value pairs, stored on the call

Values are matched case-insensitively. A value the model doesn’t list returns 400 unsupported_value with the allowed values; text over the limit returns 400 input_too_long. Other OpenAI fields, such as speed and instructions, are not supported and return 400 unsupported_parameter.

Voices

Voices, languages, formats and the length limit are in the model’s generation_config:

curl "https://api.geekidea.ai/v1/models?type=audio" -H "Authorization: Bearer $GEEKIDEA_API_KEY"

"generation_config": {
  "voice": {
    "type": "string",
    "default": "Cherry",
    "enum": ["Cherry", "Ethan", "Dylan", "Sunny", "Eric", "Rocky", "Kiki", "Peter", "Marcus", "Roy"],
    "options": [
      { "value": "Cherry", "dialect": "mandarin", "gender": "female", "name_i18n": { "en-US": "Cherry", "zh-CN": "芊悦" } },
      …
    ]
  },
  "language": { "type": "string", "enum": ["Auto", "Chinese", "English", "German", "Italian", "Portuguese", "Spanish", "Japanese", "Korean", "French", "Russian"] },
  "response_format": { "type": "string", "enum": ["wav"], "default": "wav" },
  "max_input_characters": { "type": "integer", "max": 10000 }
}

voice.options describes each voice: its dialect, gender and display names.

Look up a call

GET /v1/generations/{id} with the x-geekidea-id value returns the call with its usage and a signed link to the audio:

"output": {
  "audios": [{ "url": "https://….wav", "seconds": 4.2, "mime": "audio/wav" }]
},
"usage": { "characters": 33, "audio_seconds": 4.2, "cost": 0.00033, "currency": "USD", "cost_details": { "audio": 0.00033 } }

Links expire. Query the call again for a fresh link.

Pricing

Speech is priced per 1,000 characters: cost = price × billed characters / 1,000. The billed count is the one reported by the model provider, which may count a Chinese character as two; when the provider doesn’t report it, the number of characters in input is used. The hold assumes twice the input length, and you are charged the actual cost. See Billing.

Get started

Generate

Workflows

Async and webhooks

Account

Parameters

Voices

Look up a call

Pricing

​Parameters

​Voices

​Look up a call

​Pricing

Parameters

Voices

Look up a call

Pricing