Alibaba Cloud Model Studio:Non-real-time speech recognition (Qwen-Audio-3.0-ASR-Flash/Fun-ASR-Flash) HarmonyOS SDK

Yes

The service endpoint. Set this parameter to wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation. Replace {WorkspaceId} with your actual Workspace ID.

Yes

The API key. We recommend using a short-lived, more secure temporary API key to reduce the risk of exposing a long-lived key.

Yes

The operating mode. For non-real-time speech recognition, set this parameter to "1".

Yes

A unique string that identifies the end user. You can use an in-app user ID or a client-generated unique device identifier. This ID is mainly used for log tracing and troubleshooting.

No

The path for log files. This parameter takes effect only when you call initializeFileTrans with save_log set to true. In this case, you must specify the log file path. Otherwise, an error occurs. Up to two log files are retained locally.

No

The maximum log file size in bytes. This parameter takes effect only when you call initializeFileTrans with save_log set to true. Default: 104857600 (100 * 1024 * 1024 bytes, or 100 MiB).

No

If Connection and control parameters apikey uses temporary API key, you can update it here to prevent expiration.

Yes

The core speech recognition configuration object, including the model and parameters that control recognition.

Yes

The model name. Qwen-Audio-3.0-ASR-Flash and Fun-ASR-Flash models are supported. For details, see Supported models and regions.

Yes

The audio format. Set this parameter to the actual format. Supported formats include wav, mp3, opus. For details, see Audio specifications.

No

The audio sample rate in Hz. For example, 16000 indicates a sample rate of 16 kHz. For details, see Audio specifications.

No

The precompiled vocabulary ID. Create a vocabulary in advance and pass its ID during recognition to use its hotwords. Use this option when the vocabulary is known, relatively stable, and reused across requests. For usage instructions, see Precompiled hotwords.

No

Instant hotwords.
Pass key-value pairs. The key is the hotword text (string) and the value is its weight (integer). You do not need to create a vocabulary in advance. Valid weights are [1, 5] or 50. Within [1, 5], a higher weight makes the model more likely to output the word. A weight of 50 specifies a super hotword and significantly increases recall. You can specify up to 50 super hotwords.
Use this option for temporary, session-level hotword optimization.
When both instant and precompiled hotwords are configured, the system merges them. If the merged list exceeds 2000 entries, it randomly selects 2000 entries. For usage instructions, see Instant hotwords.

{
 "张三": 5,
 "李四": 5
}

No

The language codes for recognition. If the language is unknown in advance, omit this parameter to let the model detect it automatically.
For Qwen-Audio-3.0-ASR-Flash models, up to four values are supported. If more are specified, only the first four take effect. For Fun-ASR-Flash models, only one value is supported. If multiple values are specified, only the first takes effect.

• qwen-audio-3.0-asr-flash, fun-asr-flash-2026-06-15:
• zh: Chinese
• en: English
• ja: Japanese
• ko: Korean
• vi: Vietnamese
• th: Thai
• id: Indonesian
• ms: Malay
• tl: Filipino
• hi: Hindi
• ar: Arabic
• fr: French
• de: German
• es: Spanish
• pt: Portuguese
• ru: Russian
• it: Italian
• nl: Dutch
• sv: Swedish
• da: Danish
• fi: Finnish
• no: Norwegian
• el: Greek
• pl: Polish
• cs: Czech
• hu: Hungarian
• ro: Romanian
• bg: Bulgarian
• hr: Croatian
• sk: Slovak

messages

Yes

The list of messages. It contains the audio to recognize and optional conversation context to improve recognition.
See the following descriptions.

role

Yes

The message role. Valid values:

• user (required): A user message. When type is input_audio, it represents the audio to recognize. When type is input_text, it represents recognition results from previous turns or a domain-specific vocabulary (optional context).
• assistant (optional context): Large language model responses from previous turns.

content

Yes

The list of message content. See the following descriptions.

type

Yes

The content type. Each request requires at least one input_audio message. Valid values:

• input_audio (required): The audio input to recognize (role is user). You must also provide the input_audio object.
• input_text (optional context): Recognition results of user speech from previous turns or a domain-specific vocabulary (role is user). You must also provide the text field.
• text (optional context): Large language model responses from previous turns (role is assistant). You must also provide the text field.

input_audio

No

Required when type is input_audio.

input_audio.data

string

Yes

The audio data to recognize. For supported audio formats, file size limits, duration limits, and other input requirements, see Audio specifications. The following two input methods are supported:

• Audio file URL: Pass a publicly accessible audio file URL.
• Base64 Data URI: Pass Base64-encoded audio as a Data URI. Concatenate the data:{MIME_TYPE};base64, prefix with the Base64-encoded audio data. Supported MIME types include audio/wav, audio/mp3.
  Example (URL): https://example.com/audio/sample.wav
  Example (Base64): data:audio/wav;base64,{BASE64_ENCODED_DATA}

text

No

Required when type is input_text, provide recognition results of user speech from previous turns or a domain-specific vocabulary. When type is text, provide large language model responses from previous turns. Text length is counted in characters, with each character counting as one. The total length of all messages' text fields in each context turn must not exceed 400 characters. Excess characters are truncated from the end.

The implementation of the file transcription event and data callback interface.

A JSON string containing authentication, connection, and debugging parameters. See Connection and control parameters.

The SDK log level. Use a value from the enumeration.

Whether to save logs locally. If set to true, use Connection and control parameters to set debug_path to specify the path, and use max_log_file_size to set the file size.

Recognition parameters in nls_config parameters; parameters outside nls_config cannot be set with this method.

Recognition parameters.
Example:

{
 "apikey": "st-****",
 "messages": [
 {
 "content": [
 {
 "input_audio": {
 "data": "{YOUR_AUDIO_URL}"
 },
 "type": "input_audio"
 }
 ],
 "role": "user"
 }
 ],
 "nls_config": {
 "format": "mp3",
 "model": "qwen-audio-3.0-asr-flash"
 }
}

The task ID buffer. The SDK writes an internally generated random task ID string to this buffer. Its byte length must be >= 33 bytes (the example uses new ArrayBuffer(64)). After a successful call, decode the buffer to obtain the task's task_id.

The ID of the task to query (written to the buffer by startFileTranscriber ).

The ID of the task to cancel.

The callback event.

Error code, valid when an EVENT_ASR_ERROR event occurs.

Whether the task has finished.

The speech recognition result.

The task ID.

EVENT_FILE_TRANS_CONNECTED

Successfully connected to the service.

EVENT_FILE_TRANS_UPLOADED

The audio file was uploaded successfully.

EVENT_FILE_TRANS_QUERY_RESULT

The task query result.

EVENT_FILE_TRANS_RESULT

The final recognition result.

EVENT_ASR_ERROR

An error occurred during speech recognition.