All Products
Search
Document Center

Vector Retrieval Service for Milvus:Audio transcription

Last Updated:Oct 19, 2026

The AI_AUDIO_TRANSCRIBE function transcribes audio into text. It is suitable for scenarios such as customer service recordings, meeting minutes, voice messages, and podcast content indexing.

Command syntax

REST API

{
  "model_name": "qwen3-asr-flash",
  "texts": ["<audio_url>"],
  "params": {
    "language": "zh",
    "enable_itn": true,
    "max_concurrency": 1
  }
}

Python

schema = MilvusClient.create_schema(auto_id=True, enable_dynamic_field=False)
schema.add_field("id", DataType.INT64, is_primary=True)
schema.add_field("audio_url", DataType.VARCHAR, max_length=4096)
schema.add_field("transcript", DataType.VARCHAR, max_length=4096)
schema.add_field("dummy_vector", DataType.FLOAT_VECTOR, dim=2)
schema.add_function(
    Function(
        name="transcribe_audio",
        function_type=texttransform_function_type(),
        input_field_names=["audio_url"],
        output_field_names=["transcript"],
        params={
            "provider": "aliyun_milvus",
            "model_name": "qwen3-asr-flash",
            "task": "ai_audio_transcribe",
            "language": "zh",
            "enable_itn": "true",
        },
    )
)

Parameters

Parameter

Description

model_name

Required. Use qwen3-asr-flash.

texts

Required for REST. An array of audio URLs that must not contain empty strings. The results are returned in the same order as the inputs.

language

Optional. The source language code, for example zh or en.

enable_itn

Optional. A boolean value. It normalizes spoken forms such as spoken numbers into written expressions.

timeout_sec / max_concurrency

Optional. They control the call timeout and the number of audio files processed concurrently, respectively.

media_type

Can be omitted or set to audio. Other values cause an error.

Others

prompt, temperature, and other custom model parameters are not supported. Whether an audio URL is accessible, as well as the supported formats and durations, is subject to the limits of the model service.

Return values

data.output.outputs returns the transcribed text in one-to-one correspondence with the audio inputs. usage.audio_tokens indicates the audio usage, and usage.seconds indicates the duration of audio that has been processed.

{
  "code": 0,
  "data": {
    "output": {"outputs": ["<transcript>"]},
    "usage": {"audio_tokens": 256, "total_tokens": 256, "seconds": 45}
  }
}

Example: generate a searchable version archive for a customer service hotline welcome message

A customer service platform converts the hotline welcome message into text for version archiving and for manual review of whether content such as service hours and recording notifications is clearly expressed. Fields that do not appear in the transcript must be marked as "unconfirmed" and must not be filled in by the model. By default, the public welcome.mp3 is a Chinese welcome message, so the example uses language=zh. When you replace AIFUNC_AUDIO_URL, also set AIFUNC_AUDIO_LANGUAGE to the actual source language. The cURL example requires jq.

REST API

#!/usr/bin/env bash
set -euo pipefail

MILVUS_REST_BASE_URL="http://c-xxxx.milvus.aliyuncs.com:19530"
MILVUS_AUTH_TOKEN="<yourUsername>:<yourPassword>"

post_json() {
  local path="$1"
  local body="$2"
  curl -X POST \
    "$MILVUS_REST_BASE_URL$path" \
    -H "Authorization: Bearer $MILVUS_AUTH_TOKEN" \
    -H "Content-Type: application/json" \
    -d "$body"
}

MODEL_NAME="qwen3-asr-flash"
AUDIO_URL="${AIFUNC_AUDIO_URL:-https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3}"

BODY=$(cat <<JSON
{
  "model_name": "$MODEL_NAME",
  "texts": ["$AUDIO_URL"],
  "params": {"language": "zh", "enable_itn": true, "max_concurrency": 1}
}
JSON
)

RESPONSE_BODY="$(post_json "/v2/vectordb/ai/audio_transcribe" "$BODY")"
if command -v jq >/dev/null 2>&1; then
  echo "$RESPONSE_BODY" | jq .
  [ "$(echo "$RESPONSE_BODY" | jq -r '.code // -1')" = "0" ] || exit 1
else
  echo "$RESPONSE_BODY"
fi

Python

from __future__ import annotations

from typing import Any

from pymilvus import DataType, Function, FunctionType, MilvusClient

MILVUS_URI = "http://c-xxxx.milvus.aliyuncs.com:19530"
MILVUS_TOKEN = "<yourUsername>:<yourPassword>"

DUMMY_VECTOR_DIM = 2
TEXTTRANSFORM_FUNCTION_TYPE = 9

def texttransform_function_type() -> Any:
    for type_name in ("TEXTTRANSFORM", "TEXT_TRANSFORM", "TextTransform"):
        function_type = getattr(FunctionType, type_name, None)
        if function_type is not None:
            return function_type
    # Alibaba Cloud Milvus provides TEXTTRANSFORM as a managed extension (function type value 9);
    # some pymilvus versions do not yet include this enum member, while Function(...) validates it through FunctionType(...).
    existing = getattr(FunctionType, "_value2member_map_", {}).get(TEXTTRANSFORM_FUNCTION_TYPE)
    if existing is not None:
        return existing
    extension = int.__new__(FunctionType, TEXTTRANSFORM_FUNCTION_TYPE)
    extension._name_ = "TEXTTRANSFORM"
    extension._value_ = TEXTTRANSFORM_FUNCTION_TYPE
    FunctionType._value2member_map_[TEXTTRANSFORM_FUNCTION_TYPE] = extension
    FunctionType._member_map_["TEXTTRANSFORM"] = extension
    return extension

def add_id(schema: Any) -> None:
    schema.add_field("id", DataType.INT64, is_primary=True)

def add_dummy_vector(schema: Any) -> None:
    schema.add_field("dummy_vector", DataType.FLOAT_VECTOR, dim=DUMMY_VECTOR_DIM)

def run_texttransform_example(*, client, collection_name, input_fields, output_field, function_name, function_params, rows) -> None:
    if client.has_collection(collection_name):
        client.drop_collection(collection_name)
    schema = MilvusClient.create_schema(auto_id=True, enable_dynamic_field=False)
    add_id(schema)
    for name, data_type, max_length in input_fields:
        field_params = {"max_length": max_length} if max_length is not None else {}
        schema.add_field(name, data_type, **field_params)
    output_name, output_data_type, output_max_length = output_field
    output_params = {"max_length": output_max_length} if output_max_length is not None else {}
    schema.add_field(output_name, output_data_type, **output_params)
    add_dummy_vector(schema)
    schema.add_function(
        Function(
            name=function_name,
            function_type=texttransform_function_type(),
            input_field_names=[name for name, _, _ in input_fields],
            output_field_names=[output_name],
            params=function_params,
        )
    )
    index_params = client.prepare_index_params()
    index_params.add_index(field_name="dummy_vector", index_type="AUTOINDEX", metric_type="COSINE")
    client.create_collection(collection_name=collection_name, schema=schema, index_params=index_params)
    client.insert(collection_name, rows)
    client.flush(collection_name)
    fields = [name for name, _, _ in input_fields] + [output_name]
    for row in client.query(collection_name, filter="", output_fields=fields, limit=len(rows)):
        print(row)

MODEL_NAME = "qwen3-asr-flash"
client = MilvusClient(uri=MILVUS_URI, token=MILVUS_TOKEN)

run_texttransform_example(
    client=client,
    collection_name="simple_ai_audio_transcribe_schema",
    input_fields=[("audio_url", DataType.VARCHAR, 4096)],
    output_field=("transcript", DataType.VARCHAR, 4096),
    function_name="transcribe_audio",
    function_params={"provider": "aliyun_milvus", "model_name": MODEL_NAME, "task": "ai_audio_transcribe", "language": "zh", "enable_itn": "true"},
    rows=[{"audio_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3", "dummy_vector": [0.1, 0.2]}],
)

Expected result: data.output.outputs[0] is a non-empty welcome message transcript of no more than 10,000 characters, which can be written to the transcript field of the hotline version archive. In an actual test, the returned Chinese transcript is \u6b22\u8fce\u4e0e\u4f7f\u7528\u963f\u91cc\u4e91\u3002, which is shown here as Unicode escapes; decode them to obtain the actual Chinese characters ("Welcome to Alibaba Cloud.").

{"code": 0, "data": {"output": {"outputs": ["\u6b22\u8fce\u4e0e\u4f7f\u7528\u963f\u91cc\u4e91\u3002"]}}}

In Python scenarios, you can name the input field audio_url and the output field transcript, and let the Function perform the transcription automatically.

Before transcribing your own recordings, make sure that you have provided the required recording notification and obtained the necessary authorization. We recommend that you use short-lived media URLs with the least privilege, and limit the retention period of the original audio and the transcribed text according to your customer service data retention policy.