API Overview

Overview of API capabilities on the Unisound Maas Platform, including multi-modal capabilities like voice, vision, and file management.

Get API Key

Through Metered & Packs > API Key Management, create a new API Key to get the API Key

Text Generation

Chat with a specified model, and the model generates responses based on the request content.

Supported Models

Interface Instructions

This section includes conversational APIs aligned to different models and protocols:

Decision Models

Enterprise decision model that routes the right task to the right intelligence, returning structured judgments with probabilities and confidence.

Supported Models

Interface Instructions

This API provides structured decision calls over the SystemOne protocol:

Real-time Speech Transcription

This API receives audio streams via WebSocket and transcribes them into punctuated text in real time, suitable for live subtitles, online meetings, voice chat, intelligent assistants, and other scenarios.

The interface supports the following features:

1. Supports high-precision speech recognition for Mandarin and various dialects such as Cantonese and Sichuan dialect
2. Capable of handling complex acoustic environments, with automatic language detection and intelligent non-human voice filtering
3. Supports hotword customization to improve recognition accuracy for specific vocabulary
4. Supports speaker diarization, generating structured recognition results

Supported Models

The following are the voice models provided by Unisound and their characteristics.

Interface Instructions

This API provides stable streaming transcription over WebSocket for long-duration speech scenarios.

Async Long Audio Transcription

This API supports speech transcription based on long audio, supporting up to 5 hours of audio per request, and the generated results can be retrieved asynchronously.

The interface supports the following features:

1. Supports mp3, opus, wav, amr, m4a, ogg format audios
2. Supports speaker separation through channels or voiceprint features
3. Supports timestamp information in recognition results

Submitting a speech transcription task will generate a task_id. Use this task_id to call the retrieve task status API to get the status and result.

Applicable scenarios: Meeting transcription, customer service quality inspection, subtitle generation.

Supported Models

The following are the voice models provided by Unisound and their characteristics.

Interface Instructions

Contains 2 APIs overall: Create Async Speech Transcription Task and Query Speech Transcription Task Status. The steps are as follows:

1. Create a speech transcription task to get a task_id;
2. Query the speech transcription task status based on the task_id;
3. After the task is successful, get the speech transcription result.

Sync Speech Synthesis

This API supports synchronous text-to-speech generation, processing up to 500 characters of text per request.

The interface supports the following features:

1. Supports independent selection of system voices, cloned voices, and designed voices
2. Supports adjustment of intonation, speech rate, volume, and output format
3. Supports streaming output

Supported Models

The following are the voice models provided by Unisound and their characteristics.

Interface Instructions

This API provides synchronous text-to-speech over WebSocket with streaming audio. The steps are as follows:

1. Establish a WebSocket connection;
2. Send a text frame start with the full text to synthesize (text), and configure voice_setting / audio_setting;
3. Receive binary audio frames in order and parse text control frames;
4. When the control frame has end=true, the synthesis round is finished.

Async Long Text TTS

This API supports async text-to-speech generation, supporting a maximum of 50,000 characters per request, and the generated complete audio result can be retrieved asynchronously.

The interface supports the following features:

1. Independent selection of system voices, clone voices, and designed voices;
2. Independent adjustment of intonation, speech rate, volume, sample rate, and output format;
3. Supports inputting text to be synthesized via direct string input or uploaded text file file_id;

Submitting a long-text TTS request will generate a task_id. Use this task_id to query the synthesis task status and retrieve the result.

Applicable scenarios: Voice synthesis of long texts such as whole books.

Supported Models

The following are the voice models provided by Unisound and their characteristics.

Interface Instructions

Contains 2 APIs overall: Create Async TTS Task and Query TTS Task Status. The steps are as follows:

1. Create a TTS task to get a task_id (if you choose to pass the text via text_file_id, you need to use the File(Upload) interface first to upload the file);
2. Query the TTS task status based on the task_id;
3. If the task is generated successfully, you can use the returned file_id to view and download the result via the File API.

Fast Voice Clone

This API supports cloning voices based on uploaded audio to be cloned and sample audio.

Applicable scenarios: IP voice cloning and other scenarios requiring fast cloning of a specific voice.

Supported Models

The following are the voice models provided by Unisound and their characteristics.

Interface Instructions

1. Upload Clone Audio Call Upload Clone Audio to upload the audio file to be cloned and get the file_id.
2. Upload Prompt Audio (Optional) If prompt audio is needed to enhance the cloning effect, call Upload Prompt Audio to upload the sample audio file and get the corresponding file_id.
3. Call Clone Interface Use the obtained file_id and custom voice_id as input parameters to call Fast Clone to clone the voice.

Voice Design

This API supports generating personalized custom timbres based on user-input sound description prompts.

This API supports using the generated timbre (voice_id) in async long-text speech synthesis interfaces for speech generation.

Supported Models

The following are the voice models provided by Unisound and their characteristics.

Interface Instructions

Call Design Interface Provide a voice description prompt and optionally preview_text / custom voice_id, then call Voice Design to generate a personalized voice.

Async Document Parsing

This API supports intelligent parsing of document content, asynchronously querying parsing status and results.

The interface supports the following features:

1. Supports documents in JPG, PNG, PDF formats;
2. Supports parsing specified page numbers;

Submitting a document parsing task will generate a task_id. Use this task_id to call the retrieve task status API to get the status and result.

Applicable scenarios: Parsing content of documents such as contracts, financial reports, medical records, files, invoices, etc.

Supported Models

The following are the OCR models provided by Unisound and their characteristics.

Interface Instructions

Contains 2 APIs overall: Create Async Document Parsing Task and Query Document Parsing Task Status. The steps are as follows:

1. Create an async document parsing task to get a task_id;
2. Query the document parsing task status based on the task_id;
3. After the task is successful, get the document parsing result, including Markdown content and layout information.

Document Information Extraction and Classification

This API supports structured extraction of information from images.

Applicable scenarios: Structured data extraction from contracts, invoices, and forms.

Supported Models

The following are OCR models provided by Unisound with their feature descriptions.

API Description

Through the document information extraction and classification API, specify images and prompts to perform structured information extraction from images according to the prompt requirements.