Paraformer语音识别热词定制与管理_大模型服务平台百炼-阿里云帮助中心

Paraformer语音识别热词定制与管理

说明

支持的领域 / 任务：audio（音频） / asr（语音识别）

在语音识别服务中，如果您的业务领域有部分词汇默认识别效果不够好，可以考虑使用热词功能，将这些词添加到词表从而改善识别结果。

前提条件

已开通服务并获得API-KEY：获取API-KEY。
已安装最新版SDK：安装百炼SDK。

热词

热词通过热词列表的形式在SDK中使用，热词列表是一个以热词文本为Key，热词权重为Value的字典。热词列表最大支持设置500个热词，热词文本规则如下：纯中文热词不超过10个汉字，纯英文或者中英文混合热词，按空格分词后，不超过5个词；对于热词权重规则如下：有效的热词权重取值范围为[1, 5]和[-6, -1]区间内的整数值。如果想提高某个热词的识别概率，则可以设置[1, 5]范围内的权重，权重越大概率越高；如果想降低某个热词的识别概率，则可以设置[-6, -1]范围内的权重，权重越小概率越低。

热词管理

在Java和Python中使用AsrPhraseManager类来管理热词的创建，更新，删除，查询等功能。

导入

Python

from dashscope.audio.asr import AsrPhraseManager

Java

import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseManager;

创建热词

同步调用的形式提交一个创建热词请求。

接口

Python

def create_phrases(cls,
                   model: str,
                   phrases: Dict[str, Any],
                   training_type: str = 'compile_asr_phrase',
                   **kwargs)

Java

AsrPhraseStatusResult CreatePhrases(AsrPhraseParam param)
      throws ApiException, NoApiKeyException, InputRequiredException

参数说明

对于Java SDK，将使用一个AsrPhraseParam对象作为参数，其方法和参数如下：

参数	类型	说明
param	AsrPhraseParam	创建热词的配置参数，见上文关于AsrPhraseParam类型的描述，CreatePhrases调用不需要填写pageNo和pageSize字段，但是要求添加model，phraseList字段。

对于Python SDK，其参数说明如下：

参数	类型	说明
model	str	指定的Paraformer模型名，关于如何进行模型选择，请参考：模型概览
phrases	Dict[str, Any]	热词列表。
training_type	str	固定为compile_asr_phrase。

模型概览

模型名	模型简介
paraformer-v1	Paraformer中英文语音识别模型，支持16kHz及以上采样率的音频或视频语音识别。
paraformer-8k-v1	Paraformer中文语音识别模型，支持8kHz电话语音识别。
paraformer-mtl-v1	Paraformer多语言语音识别模型，支持16kHz及以上采样率的音频或视频语音识别。支持的语种/方言包括：中文普通话、中文方言（粤语、吴语、闽南语、东北话、甘肃话、贵州话、河南话、湖北话、湖南话、宁夏话、山西话、陕西话、山东话、四川话、天津话）、英语、日语、韩语、西班牙语、印尼语、法语、德语、意大利语、马来语。

返回示例

对于Java SDK，将返回一个AsrPhraseStatusResult对象，对于Python SDK，将返回一个Dict，AsrPhraseStatusResult成员通过对应get方法获取，成员名称和Python SDK基本一致，仅命名方式不同（Java为驼峰式）。

{
	"status_code": 200,
	"request_id": "2b815cfe-793f-9f3c-b528-5ade0a2d498e",
	"code": null,
	"message": "",
	"output": {
		"job_id": "ft-202309261539-2af1",
		"status": "SUCCEEDED",
		"finetuned_output": "paraformer-realtime-v1-ft-202309261539-2af1",
		"training_type": "compile_asr_phrase",
		"create_time": "2023-09-26 15:39:07"
	},
	"usage": null,
	"job_id": "ft-202309261539-2af1",
	"finetuned_output": "paraformer-realtime-v1-ft-202309261539-2af1",
	"finetuned_outputs": null,
	"training_type": null,
	"create_time": "2023-09-26 15:39:07",
	"output_type": null,
	"model": null
}

调用范例

Python

# coding=utf-8

import dashscope
from dashscope.audio.asr import AsrPhraseManager

dashscope.api_key='your-dashscope-api-key'

phrases = {'通义千问': 5}

result = AsrPhraseManager.create_phrases(model='paraformer-realtime-v1',
                                         phrases=phrases)
if result.output is not None and result.output['finetuned_output'] is not None:
    print('job_id:%s, finetuned_output:%s' %
          (result.output['job_id'], result.output['finetuned_output']))
else:
    print('Error: ', str(result))

Java

package com.alibaba.test;

import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseManager;
import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseParam;
import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseStatusResult;
import com.alibaba.dashscope.audio.asr.recognition.Recognition;
import com.alibaba.dashscope.audio.asr.recognition.RecognitionParam;
import com.alibaba.dashscope.utils.Constants;

class Test {
  
  public static void main(String[] args) {
    AsrPhraseParam param = AsrPhraseParam.builder()
            .model("your-model")
            .phraseList(Collections.singletonMap("通义千问", 5))
            .apiKey("your-dashscope-api-key")
            .build();

    AsrPhraseStatusResult createResult = null;
    try {
      createResult = AsrPhraseManager.CreatePhrases(param);
      if (createResult.getOutput() != null && createResult.getOutput().getFineTunedOutput() != null) {
        System.out.println("job_id: " + createResult.getOutput().getJobId() + ", finetuned_output: " + createResult.getOutput().getFineTunedOutput());
      } else {
        System.out.println("Error: " + createResult);
      }
    } catch (Exception e) {
      throw new RuntimeException(e);
    }
  }
}

查询热词

该接口将以同步调用的形式提交一个查询热词请求。

接口

Python

def query_phrases(cls, phrase_id: str, **kwargs)

Java

AsrPhraseStatusResult QueryPhrase(AsrPhraseParam param, String phraseId)
      throws ApiException, NoApiKeyException, InputRequiredException

参数配置

对于Java SDK，将使用一个AsrPhraseParam对象作为参数，其方法和参数如下：

参数	类型	说明
param	AsrPhraseParam	创建热词的配置参数，见上文关于AsrPhraseParam类型的描述，对于QueryPhrase调用，不需要填phraseLIst及pageNo和pageSize字段。
phraseId	String	调用CreatePhrases，UpdatePhrases, QueryPhrase等接口返回的AsrPhraseStatusOutput对象后，通过该对象的getFineTunedOutput返回的热词ID，String类型。调用ListPhrases时，使用AsrPhraseStatusOutput对象的getFinetunedOutputs接口将返回所有热词信息的列表，然后使用AsrPhraseInfo的getFineTunedOutput即可获取对应热词的热词ID。

对于Python SDK，参数说明如下：

参数	类型	说明
phrase_id	str	调用create_phrases，update_phrases, query_phrases等接口返回的Dict对象，通过finetuned_output访问对应phrase_id

返回示例

{
	"status_code": 200,
	"request_id": "19ee9c5f-173b-9fed-8e61-40bc53f1eea7",
	"code": null,
	"message": "",
	"output": {
		"create_time": "2023-09-26 15:39:08",
		"finetuned_output": "paraformer-realtime-v1-ft-202309261539-2af1",
		"job_id": "ft-202309261539-2af1",
		"model": "paraformer-realtime-v1",
		"output_type": "custom_resource"
	},
	"usage": null,
	"job_id": "ft-202309261539-2af1",
	"finetuned_output": "paraformer-realtime-v1-ft-202309261539-2af1",
	"finetuned_outputs": null,
	"training_type": null,
	"create_time": "2023-09-26 15:39:08",
	"output_type": "custom_resource",
	"model": "paraformer-realtime-v1"
}

调用示例

Python

# coding=utf-8

import dashscope
from dashscope.audio.asr import AsrPhraseManager

dashscope.api_key='your-dashscope-api-key'

result = AsrPhraseManager.query_phrases(phrase_id='phrase-id')
if result.output is not None:
    print('query phrases: ', result.output)
else:
    print('Error: ', str(result))

Java

package com.alibaba.test;

import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseManager;
import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseParam;
import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseStatusResult;
import com.alibaba.dashscope.audio.asr.recognition.Recognition;
import com.alibaba.dashscope.audio.asr.recognition.RecognitionParam;
import com.alibaba.dashscope.utils.Constants;

class Test {
  
  public static void main(String[] args) {
    AsrPhraseParam param = AsrPhraseParam.builder()
            .model("your-model")
            .apiKey("your-dashscope-api-key")
            .build();

    AsrPhraseStatusResult result = null;
    try {
      result = AsrPhraseManager.QueryPhrase(param, "phrase-id");
      if (result.getOutput() != null) {
        System.out.println("query phrases: " + result.getOutput());
      } else {
        System.out.println("Error: " + result);
      }
    } catch (Exception e) {
      throw new RuntimeException(e);
    }
  }
}

更新热词

该接口将以同步调用的形式提交一个更新热词请求。

接口

Python

def update_phrases(cls,
                   model: str,
                   phrase_id: str,
                   phrases: Dict[str, Any],
                   training_type: str = 'compile_asr_phrase',
                   **kwargs)

Java

AsrPhraseStatusResult UpdatePhrases(AsrPhraseParam param, String phraseId)
      throws ApiException, NoApiKeyException, InputRequiredException

参数配置

对于Java SDK，将使用一个AsrPhraseParam对象作为参数，其方法和参数如下：

参数	类型	说明
param	AsrPhraseParam	创建热词的配置参数，见上文关于AsrPhraseParam类型的描述，对于UpdatePhrase调用，不需要填pageNo和pageSize字段。
phraseId	String	调用CreatePhrases，UpdatePhrases, QueryPhrase等接口返回的AsrPhraseStatusOutput对象后，通过该对象的getFineTunedOutput返回的热词ID，String类型。调用ListPhrases时，使用AsrPhraseStatusOutput对象的getFinetunedOutputs接口将返回所有热词信息的列表，然后使用AsrPhraseInfo的getFineTunedOutput即可获取对应热词的热词ID。

对于Python SDK，其参数如下：

参数	类型	默认值	说明
model	str	-	指定用于音视频文件转写的Paraformer模型名，关于如何进行模型选择，请参考：模型概览。
phrase_id	str	-	调用create_phrases，update_phrases, query_phrases等接口返回的Dict对象，通过finetuned_output访问对应phrase_id。
phrases	Dict[str, Any]	-	热词列表，是一个Dict类型对象，其中键为热词文本，值为热词对应权重。对于热词要求请参考下方重要一栏。
training_type	str	compile_asr_phrase	固定为compile_asr_phrase

返回示例

{
	"status_code": 200,
	"request_id": "8c8d64e3-5198-9624-99cd-e9dcf7eb22f6",
	"code": null,
	"message": "",
	"output": {
		"job_id": "ft-202309261543-b0ae",
		"status": "SUCCEEDED",
		"finetuned_output": "paraformer-realtime-v1-ft-202309261539-2af1",
		"training_type": "compile_asr_phrase",
		"create_time": "2023-09-26 15:43:09"
	},
	"usage": null,
	"job_id": "ft-202309261543-b0ae",
	"finetuned_output": "paraformer-realtime-v1-ft-202309261539-2af1",
	"finetuned_outputs": null,
	"training_type": null,
	"create_time": "2023-09-26 15:43:09",
	"output_type": null,
	"model": null
}

调用示例

Python

# coding=utf-8

import dashscope
from dashscope.audio.asr import AsrPhraseManager

dashscope.api_key='your-dashscope-api-key'

phrases = {'通义千问': 2}

result = AsrPhraseManager.update_phrases(model='paraformer-realtime-v1',
                                         phrase_id='phrase-id',
                                         phrases=phrases)
if result.output is not None:
    print('update phrases: ', result.output)
else:
    print('Error: ', str(result))

Java

package com.alibaba.test;

import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseManager;
import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseParam;
import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseStatusResult;
import com.alibaba.dashscope.audio.asr.recognition.Recognition;
import com.alibaba.dashscope.audio.asr.recognition.RecognitionParam;
import com.alibaba.dashscope.utils.Constants;

class Test {
  
  public static void main(String[] args) {
    AsrPhraseParam param = AsrPhraseParam.builder()
      .model("your-model")
      .phraseList(Collections.singletonMap("通义千问", 2))
      .apiKey("your-dashscope-api-key")
      .build();

    AsrPhraseStatusResult result = null;
    try {
      result = AsrPhraseManager.UpdatePhrases(param, "phrase-id");
      if (result.getOutput() != null) {
        System.out.println("update phrases: " + result.getOutput());
      } else {
        System.out.println("err: " + result);
      }
    } catch (Exception e) {
      throw new RuntimeException(e);
    }
  }
}

删除热词

该接口将以同步调用的形式提交一个热词删除请求。

接口

Python

def delete_phrases(cls, phrase_id: str,
                   **kwargs)

Java

AsrPhraseStatusResult DeletePhrase(AsrPhraseParam param, String phraseId)
      throws ApiException, NoApiKeyException, InputRequiredException

参数配置

对于Java SDK，将使用一个AsrPhraseParam对象作为参数，其方法和参数如下：

参数	类型	说明
param	AsrPhraseParam	创建热词的配置参数，见上文关于AsrPhraseParam类型的描述，对于DeletePhrase调用，不需要填phraseList, pageNo和pageSize字段。
phraseId	String	调用CreatePhrases，UpdatePhrases, QueryPhrase等接口返回的AsrPhraseStatusOutput对象后，通过该对象的getFineTunedOutput返回的热词ID，String类型。调用ListPhrases时，使用AsrPhraseStatusOutput对象的getFinetunedOutputs接口将返回所有热词信息的列表，然后使用AsrPhraseInfo的getFineTunedOutput即可获取对应热词的热词ID。

对于Python SDK，其参数如下：

参数	类型	默认值	说明
phrase_id	str	-	调用create_phrases，update_phrases, query_phrases等接口返回的Dict对象，通过finetuned_output访问对应phrase_id。

返回示例

{
	"status_code": 200,
	"request_id": "00bb0287-2593-94a3-8e21-93c90f5e9dd8",
	"code": null,
	"message": "",
	"output": {
		"finetuned_output": "paraformer-realtime-v1-ft-202309261539-2af1"
	},
	"usage": null,
	"job_id": null,
	"finetuned_output": "paraformer-realtime-v1-ft-202309261539-2af1",
	"finetuned_outputs": null,
	"training_type": null,
	"create_time": null,
	"output_type": null,
	"model": null
}

调用示例

Python

# coding=utf-8

import dashscope
from dashscope.audio.asr import AsrPhraseManager

dashscope.api_key='your-dashscope-api-key'

result = AsrPhraseManager.delete_phrases(phrase_id='phrase-id')
if result.output is not None:
    print('delete phrases: ', result.output)
else:
    print('Error: ', str(result))

Java

package com.alibaba.test;

import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseManager;
import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseParam;
import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseStatusResult;
import com.alibaba.dashscope.audio.asr.recognition.Recognition;
import com.alibaba.dashscope.audio.asr.recognition.RecognitionParam;
import com.alibaba.dashscope.utils.Constants;

class Test {
  
  public static void main(String[] args) {
    AsrPhraseParam param = AsrPhraseParam.builder()
            .model("your-model")
            .apiKey("your-dashscope-api-key")
            .build();

    AsrPhraseStatusResult result = null;
    try {
      result = AsrPhraseManager.DeletePhrase(param, "phrase-id");
      if (result.getOutput() != null) {
        System.out.println("delete phrases: " + result.getOutput());
      } else {
        System.out.println("Error: " + result);
      }
    } catch (Exception e) {
      throw new RuntimeException(e);
    }
  }
}

列表形式返回所有热词

该接口将以同步调用的形式提交一个返回所有热词的请求。

接口

Python

def list_phrases(cls,
                 page: int = 1,
                 page_size: int = 10,
                 **kwargs)

Java

AsrPhraseStatusResult ListPhrases(AsrPhraseParam param)
      throws ApiException, NoApiKeyException, InputRequiredException

参数配置

对于Java SDK，将使用一个AsrPhraseParam对象作为参数，其方法和参数如下：

参数	类型	说明
param	AsrPhraseParam	创建热词的配置参数，见上文关于AsrPhraseParam类型的描述，对于ListPhrases调用，不需要填phraseList字段。
phraseId	String	调用CreatePhrases，UpdatePhrases, QueryPhrase等接口返回的AsrPhraseStatusOutput对象后，通过该对象的getFineTunedOutput返回的热词ID，String类型。调用ListPhrases时，使用AsrPhraseStatusOutput对象的getFinetunedOutputs接口将返回所有热词信息的列表，然后使用AsrPhraseInfo的getFineTunedOutput即可获取对应热词的热词ID。

对于Python SDK，其参数如下：

参数	类型	默认值	说明
page	int	1	当请求是list_phrases有效，用于查询第几页列表，默认1
page_size	int	10	当请求是list_phrases有效，用于设置分页大小，默认10

返回示例

{
	"status_code": 200,
	"request_id": "95f969ef-bcbc-9bb5-b05d-4caed9326409",
	"code": null,
	"message": "",
	"output": {
		"page_no": 1,
		"page_size": 5,
		"total": 61,
		"finetuned_outputs": [{
			"create_time": "2023-09-26 15:32:20",
			"finetuned_output": "paraformer-realtime-v1-ft-202309261532-93bf",
			"job_id": "ft-202309261532-93bf",
			"model": "paraformer-realtime-v1",
			"output_type": "custom_resource"
		}, {
			"create_time": "2023-09-26 15:32:18",
			"finetuned_output": "paraformer-realtime-v1-ft-202309261532-7b51",
			"job_id": "ft-202309261532-7b51",
			"model": "paraformer-realtime-v1",
			"output_type": "custom_resource"
		}, {
			"create_time": "2023-09-26 15:32:17",
			"finetuned_output": "paraformer-realtime-v1-ft-202309261532-8bbc",
			"job_id": "ft-202309261532-cef0",
			"model": "paraformer-realtime-v1",
			"output_type": "custom_resource"
		}, {
			"create_time": "2023-09-26 15:32:16",
			"finetuned_output": "paraformer-realtime-v1-ft-202309261532-fc6a",
			"job_id": "ft-202309261532-fc6a",
			"model": "paraformer-realtime-v1",
			"output_type": "custom_resource"
		}, {
			"create_time": "2023-09-26 15:31:56",
			"finetuned_output": "paraformer-realtime-v1-ft-202309261531-e92d",
			"job_id": "ft-202309261531-e92d",
			"model": "paraformer-realtime-v1",
			"output_type": "custom_resource"
		}]
	},
	"usage": null,
	"job_id": null,
	"finetuned_output": null,
	"finetuned_outputs": [{
		"create_time": "2023-09-26 15:32:20",
		"finetuned_output": "paraformer-realtime-v1-ft-202309261532-93bf",
		"job_id": "ft-202309261532-93bf",
		"model": "paraformer-realtime-v1",
		"output_type": "custom_resource"
	}, {
		"create_time": "2023-09-26 15:32:18",
		"finetuned_output": "paraformer-realtime-v1-ft-202309261532-7b51",
		"job_id": "ft-202309261532-7b51",
		"model": "paraformer-realtime-v1",
		"output_type": "custom_resource"
	}, {
		"create_time": "2023-09-26 15:32:17",
		"finetuned_output": "paraformer-realtime-v1-ft-202309261532-8bbc",
		"job_id": "ft-202309261532-cef0",
		"model": "paraformer-realtime-v1",
		"output_type": "custom_resource"
	}, {
		"create_time": "2023-09-26 15:32:16",
		"finetuned_output": "paraformer-realtime-v1-ft-202309261532-fc6a",
		"job_id": "ft-202309261532-fc6a",
		"model": "paraformer-realtime-v1",
		"output_type": "custom_resource"
	}, {
		"create_time": "2023-09-26 15:31:56",
		"finetuned_output": "paraformer-realtime-v1-ft-202309261531-e92d",
		"job_id": "ft-202309261531-e92d",
		"model": "paraformer-realtime-v1",
		"output_type": "custom_resource"
	}],
	"training_type": null,
	"create_time": null,
	"output_type": null,
	"model": null
}

调用示例

Python

# coding=utf-8

import dashscope
from dashscope.audio.asr import AsrPhraseManager

dashscope.api_key='your-dashscope-api-key'

result = AsrPhraseManager.list_phrases()   
if result.output is not None:
    print('update phrases: ', result.output)
else:
    print('Error: ', str(result))

Java

package com.alibaba.test;

import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseManager;
import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseParam;
import com.alibaba.dashscope.audio.asr.phrase.AsrPhraseStatusResult;
import com.alibaba.dashscope.audio.asr.recognition.Recognition;
import com.alibaba.dashscope.audio.asr.recognition.RecognitionParam;
import com.alibaba.dashscope.utils.Constants;

class Test {
  
  public void static main(String[] args) {
    AsrPhraseParam param = AsrPhraseParam.builder()
            .model("your-model")
            .apiKey("your-dashscope-api-key")
            .build();

    AsrPhraseStatusResult result = null;
    try {
      result = AsrPhraseManager.ListPhrases(param);
      if (result.getOutput() != null) {
        System.out.println("list phrases: " + result.getOutput());
      } else {
        System.out.println("Error: " + result);
      }
    } catch (Exception e) {
      throw new RuntimeException(e);
    }
  }
}