Cheetah Model API

The Cheetah Model API trains a Speech-to-Text model from a YAML content definition. Send a YAML content as a request and receive a trained .pv model file in response, ready for on-device inference with Cheetah.

Base URL

All routes are relative to the following base URL:

https://rest.picovoice.ai

Authentication

Every request must be authenticated with your AccessKey passed in the x-api-key header. You can obtain a free AccessKey from the Picovoice Console.

x-api-key: ${ACCESS_KEY}

Requests made without a valid AccessKey are rejected with a 403 Forbidden response.

Train a Model

POST /{language}/api/cat

Trains a Cheetah model for the given language, model_type and target engine from a YAML definition. On success, the response body is the binary .rhn model file.

Path Parameters

• language string — The language of the context. Must be one of the supported language values (e.g. en).

Headers

• x-api-key string — Required. Your AccessKey from the Picovoice Console.

Body Parameters

The request body is a JSON object with the following fields:

• engine string — Required. The target engine to train the model for. Must be one of the supported engine values. For Cheetah Speech-to-Text use cheetah.
• model_type string — Required. The target model variant. Must be one of the supported model type values.
• yaml_content string — Required. The content definition, as a YAML string. Follow the format as exported from the Picovoice Console.

Example Request

A successful request returns a 303 See Other redirect pointing to the trained model. The -L flag tells curl to follow that redirect automatically and download the .pv file:

curl -L \
  -H "x-api-key: ${ACCESS_KEY}" \
  -o "${OUTPUT_PATH}" \
  -d '{"engine": "${ENGINE}", "model_type": "${MODEL_TYPE}", "yaml_content": "${YAML_STRING}"}' \
  "https://rest.picovoice.ai/${LANGUAGE}/api/cat"

Replace the placeholders with your own values:

• ${ACCESS_KEY} — your AccessKey from the Picovoice Console.
• ${OUTPUT_PATH} — the local path to write the returned .pv model to (e.g. ./cheetah_params.pv).
• ${ENGINE} — a supported engine value (e.g. cheetah).
• ${MODEL_TYPE} — a supported model type value (e.g. cheetah).
• ${YAML_STRING} — your content definition as a YAML string.
• ${LANGUAGE} — a supported language value (e.g. en).

Responses

• 200 OK — Returned when the 303 redirect is followed. The response body is the binary .pv model file.
• 303 See Other — Training succeeded. The response carries a Location header with a URL to the trained model. Follow this redirect (e.g. with curl -L) to download it and receive the 200 OK above.
• 400 Bad Request — The request was malformed, the engine/model_type value was invalid, or the yaml_content was not valid YAML / used an incorrect content format.
• 403 Forbidden — The AccessKey is missing or invalid.

Supported Languages

The language path parameter accepts one of the following values:

Supported Engines

The engine body parameter accepts one of the following values:

Supported Model Types

The model_type body parameter accepts one of the following values: