使用发音评估

在本文中，你将了解如何通过语音 SDK 使用语音转文本来评估发音。 发音评估可以评估语音发音，并为说话人提供有关讲话音频准确度和流利度的反馈。

在流式处理模式下使用发音评估

发音评估支持不间断的流式处理模式。 通过使用语音 SDK，录制时间可以不受限制。 只要不停止录制，评估过程就不会结束，你可以方便地暂停和恢复评估。

有关发音评估的可用性的信息，请参阅 支持的语言和可用性区域。

基线是，对于即用即付定价，发音评估的使用费率与语音转文本的费用相同。 有关详细信息，请参阅定价。

连续识别

设置配置参数

在 SpeechRecognizer 中，可以指定要学习或练习以改进发音的语言。 默认区域设置为 en-US。 要了解如何在自己的应用程序中指定用于发音评估的学习语言，可以使用以下示例代码。

var recognizer = new SpeechRecognizer(speechConfig, "en-US", audioConfig);

auto recognizer = SpeechRecognizer::FromConfig(speechConfig, "en-US", audioConfig);

SpeechRecognizer recognizer = new SpeechRecognizer(speechConfig, "en-US", audioConfig);

speech_recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, language="en-US", audio_config=audio_config)

speechConfig.speechRecognitionLanguage = "en-US";

SPXSpeechRecognizer* recognizer = [[SPXSpeechRecognizer alloc] initWithSpeechConfiguration:speechConfig language:@"en-US" audioConfiguration:audioConfig];

let recognizer = try! SPXSpeechRecognizer(speechConfiguration: speechConfig, language: "en-US", audioConfiguration: audioConfig)

必须创建PronunciationAssessmentConfig对象。 可以设置 EnableProsodyAssessment 和 EnableContentAssessmentWithTopic 以启用韵律评估和内容评估。 有关详细信息，请参阅配置方法。

var pronunciationAssessmentConfig = new PronunciationAssessmentConfig( 
    referenceText: "", 
    gradingSystem: GradingSystem.HundredMark,  
    granularity: Granularity.Phoneme,  
    enableMiscue: false); 
pronunciationAssessmentConfig.EnableProsodyAssessment(); 
pronunciationAssessmentConfig.EnableContentAssessmentWithTopic("greeting"); 

auto pronunciationConfig = PronunciationAssessmentConfig::Create("", PronunciationAssessmentGradingSystem::HundredMark, PronunciationAssessmentGranularity::Phoneme, false); 
pronunciationConfig->EnableProsodyAssessment(); 
pronunciationConfig->EnableContentAssessmentWithTopic("greeting"); 

PronunciationAssessmentConfig pronunciationConfig = new PronunciationAssessmentConfig("", 
    PronunciationAssessmentGradingSystem.HundredMark, PronunciationAssessmentGranularity.Phoneme, false); 
pronunciationConfig.enableProsodyAssessment(); 
pronunciationConfig.enableContentAssessmentWithTopic("greeting");

pronunciation_config = speechsdk.PronunciationAssessmentConfig( 
    reference_text="", 
    grading_system=speechsdk.PronunciationAssessmentGradingSystem.HundredMark, 
    granularity=speechsdk.PronunciationAssessmentGranularity.Phoneme, 
    enable_miscue=False) 
pronunciation_config.enable_prosody_assessment() 
pronunciation_config.enable_content_assessment_with_topic("greeting")

var pronunciationAssessmentConfig = new sdk.PronunciationAssessmentConfig( 
    referenceText: "", 
    gradingSystem: sdk.PronunciationAssessmentGradingSystem.HundredMark,  
    granularity: sdk.PronunciationAssessmentGranularity.Phoneme,  
    enableMiscue: false); 
pronunciationAssessmentConfig.enableProsodyAssessment(); 
pronunciationAssessmentConfig.enableContentAssessmentWithTopic("greeting");  

SPXPronunciationAssessmentConfiguration *pronunicationConfig = 
[[SPXPronunciationAssessmentConfiguration alloc] init:@"" gradingSystem:SPXPronunciationAssessmentGradingSystem_HundredMark granularity:SPXPronunciationAssessmentGranularity_Phoneme enableMiscue:false]; 
[pronunicationConfig enableProsodyAssessment]; 
[pronunicationConfig enableContentAssessmentWithTopic:@"greeting"]; 

let pronAssessmentConfig = try! SPXPronunciationAssessmentConfiguration("", 
    gradingSystem: .hundredMark, 
    granularity: .phoneme, 
    enableMiscue: false) 
pronAssessmentConfig.enableProsodyAssessment() 
pronAssessmentConfig.enableContentAssessment(withTopic: "greeting")

下表列出了发音评估的部分关键配置参数。

配置方法

下表列出了可为 PronunciationAssessmentConfig 对象设置的一些可选方法。

获取发音评估结果

当语音被识别时，你可以将发音评估结果请求为 SDK 对象或 JSON 字符串。

using (var speechRecognizer = new SpeechRecognizer(
    speechConfig,
    audioConfig))
{
    // (Optional) get the session ID
    speechRecognizer.SessionStarted += (s, e) => {
        Console.WriteLine($"SESSION ID: {e.SessionId}");
    };
    pronunciationAssessmentConfig.ApplyTo(speechRecognizer);
    var speechRecognitionResult = await speechRecognizer.RecognizeOnceAsync();

    // The pronunciation assessment result as a Speech SDK object
    var pronunciationAssessmentResult =
        PronunciationAssessmentResult.FromResult(speechRecognitionResult);

    // The pronunciation assessment result as a JSON string
    var pronunciationAssessmentResultJson = speechRecognitionResult.Properties.GetProperty(PropertyId.SpeechServiceResponse_JsonResult);
}

auto speechRecognizer = SpeechRecognizer::FromConfig(
    speechConfig,
    audioConfig);
// (Optional) get the session ID
speechRecognizer->SessionStarted.Connect([](const SessionEventArgs& e) {
    std::cout << "SESSION ID: " << e.SessionId << std::endl;
});
pronunciationAssessmentConfig->ApplyTo(speechRecognizer);
speechRecognitionResult = speechRecognizer->RecognizeOnceAsync().get();

// The pronunciation assessment result as a Speech SDK object
auto pronunciationAssessmentResult =
    PronunciationAssessmentResult::FromResult(speechRecognitionResult);

// The pronunciation assessment result as a JSON string
auto pronunciationAssessmentResultJson = speechRecognitionResult->Properties.GetProperty(PropertyId::SpeechServiceResponse_JsonResult);

SpeechRecognizer speechRecognizer = new SpeechRecognizer(
    speechConfig,
    audioConfig);
// (Optional) get the session ID
speechRecognizer.sessionStarted.addEventListener((s, e) -> {
    System.out.println("SESSION ID: " + e.getSessionId());
});
pronunciationAssessmentConfig.applyTo(speechRecognizer);
Future<SpeechRecognitionResult> future = speechRecognizer.recognizeOnceAsync();
SpeechRecognitionResult speechRecognitionResult = future.get(30, TimeUnit.SECONDS);

// The pronunciation assessment result as a Speech SDK object
PronunciationAssessmentResult pronunciationAssessmentResult =
    PronunciationAssessmentResult.fromResult(speechRecognitionResult);

// The pronunciation assessment result as a JSON string
String pronunciationAssessmentResultJson = speechRecognitionResult.getProperties().getProperty(PropertyId.SpeechServiceResponse_JsonResult);

recognizer.close();
speechConfig.close();
audioConfig.close();
pronunciationAssessmentConfig.close();
speechRecognitionResult.close();

var speechRecognizer = SpeechSDK.SpeechRecognizer.FromConfig(speechConfig, audioConfig);
// (Optional) get the session ID
speechRecognizer.sessionStarted = (s, e) => {
    console.log(`SESSION ID: ${e.sessionId}`);
};
pronunciationAssessmentConfig.applyTo(speechRecognizer);

speechRecognizer.recognizeOnceAsync((speechRecognitionResult: SpeechSDK.SpeechRecognitionResult) => {
    // The pronunciation assessment result as a Speech SDK object
    var pronunciationAssessmentResult = SpeechSDK.PronunciationAssessmentResult.fromResult(speechRecognitionResult);

    // The pronunciation assessment result as a JSON string
    var pronunciationAssessmentResultJson = speechRecognitionResult.properties.getProperty(SpeechSDK.PropertyId.SpeechServiceResponse_JsonResult);
},
{});

speech_recognizer = speechsdk.SpeechRecognizer(
        speech_config=speech_config, \
        audio_config=audio_config)
# (Optional) get the session ID
speech_recognizer.session_started.connect(lambda evt: print(f"SESSION ID: {evt.session_id}"))
pronunciation_assessment_config.apply_to(speech_recognizer)
speech_recognition_result = speech_recognizer.recognize_once()
# The pronunciation assessment result as a Speech SDK object
pronunciation_assessment_result = speechsdk.PronunciationAssessmentResult(speech_recognition_result)

# The pronunciation assessment result as a JSON string
pronunciation_assessment_result_json = speech_recognition_result.properties.get(speechsdk.PropertyId.SpeechServiceResponse_JsonResult)

SPXSpeechRecognizer* speechRecognizer = \
        [[SPXSpeechRecognizer alloc] initWithSpeechConfiguration:speechConfig
                                              audioConfiguration:audioConfig];
// (Optional) get the session ID
[speechRecognizer addSessionStartedEventHandler: ^ (SPXRecognizer *sender, SPXSessionEventArgs *eventArgs) {
    NSLog(@"SESSION ID: %@", eventArgs.sessionId);
}];
[pronunciationAssessmentConfig applyToRecognizer:speechRecognizer];

SPXSpeechRecognitionResult *speechRecognitionResult = [speechRecognizer recognizeOnce];

// The pronunciation assessment result as a Speech SDK object
SPXPronunciationAssessmentResult* pronunciationAssessmentResult = [[SPXPronunciationAssessmentResult alloc] init:speechRecognitionResult];

// The pronunciation assessment result as a JSON string
NSString* pronunciationAssessmentResultJson = [speechRecognitionResult.properties getPropertyByName:SPXSpeechServiceResponseJsonResult];

let speechRecognizer = try! SPXSpeechRecognizer(speechConfiguration: speechConfig, audioConfiguration: audioConfig)
// (Optional) get the session ID
speechRecognizer.addSessionStartedEventHandler { (sender, evt) in
	print("SESSION ID: \(evt.sessionId)")
try! pronConfig.apply(to: speechRecognizer)

let speechRecognitionResult = try? speechRecognizer.recognizeOnce()

// The pronunciation assessment result as a Speech SDK object
let pronunciationAssessmentResult = SPXPronunciationAssessmentResult(speechRecognitionResult!)

// The pronunciation assessment result as a JSON string
let pronunciationAssessmentResultJson = speechRecognitionResult!.properties?.getPropertyBy(SPXPropertyId.speechServiceResponseJsonResult)

结果参数

根据使用的是脚本化还是未脚本化评估，可以获取不同的发音评估结果。 脚本化评估适用于阅读语言学习场景。 非脚本化评估适用于口语学习场景。

脚本化评估结果

下表列出了脚本化评估或阅读场景的一些关键发音评估结果。

未脚本化评估结果

下表列出了非脚本化评估或口语场景的一些关键发音评估结果。

VocabularyScore、GrammarScore 和 TopicScore 参数汇总到合并的内容评估中。

下表更详细地描述了韵律评估结果：

JSON 结果示例

语音单词“hello”的脚本化发音评估结果显示为以下示例中的 JSON 字符串。

• 音素字母是 IPA。
• 音节与相同单词的音素一起返回。
• 可使用 Offset 和 Duration 值将音节与其对应的音素对齐。 例如，第二个音节 loʊ 的起始偏移量 (11700000) 与第三个音素 l 对齐。 偏移量表示已识别的语音在音频流中开始的时间。 此值以 100 纳秒为单位衡量。 若要详细了解 Offset 和 Duration，请参阅响应属性。
• 有五个 NBestPhonemes 对应于请求的语音音素的数量。
• 在 Phonemes 中，最可能的口语音素是 ə，而不是预期的音素 ɛ。 预期的音素 ɛ 仅获得 47 分的置信度。 其他潜在匹配的置信度分数为 52、17 和 2。

{
    "Id": "bbb42ea51bdb46d19a1d685e635fe173",
    "RecognitionStatus": 0,
    "Offset": 7500000,
    "Duration": 13800000,
    "DisplayText": "Hello.",
    "NBest": [
        {
            "Confidence": 0.975003,
            "Lexical": "hello",
            "ITN": "hello",
            "MaskedITN": "hello",
            "Display": "Hello.",
            "PronunciationAssessment": {
                "AccuracyScore": 100,
                "FluencyScore": 100,
                "CompletenessScore": 100,
                "PronScore": 100
            },
            "Words": [
                {
                    "Word": "hello",
                    "Offset": 7500000,
                    "Duration": 13800000,
                    "PronunciationAssessment": {
                        "AccuracyScore": 99.0,
                        "ErrorType": "None"
                    },
                    "Syllables": [
                        {
                            "Syllable": "hɛ",
                            "PronunciationAssessment": {
                                "AccuracyScore": 91.0
                            },
                            "Offset": 7500000,
                            "Duration": 4100000
                        },
                        {
                            "Syllable": "loʊ",
                            "PronunciationAssessment": {
                                "AccuracyScore": 100.0
                            },
                            "Offset": 11700000,
                            "Duration": 9600000
                        }
                    ],
                    "Phonemes": [
                        {
                            "Phoneme": "h",
                            "PronunciationAssessment": {
                                "AccuracyScore": 98.0,
                                "NBestPhonemes": [
                                    {
                                        "Phoneme": "h",
                                        "Score": 100.0
                                    },
                                    {
                                        "Phoneme": "oʊ",
                                        "Score": 52.0
                                    },
                                    {
                                        "Phoneme": "ə",
                                        "Score": 35.0
                                    },
                                    {
                                        "Phoneme": "k",
                                        "Score": 23.0
                                    },
                                    {
                                        "Phoneme": "æ",
                                        "Score": 20.0
                                    }
                                ]
                            },
                            "Offset": 7500000,
                            "Duration": 3500000
                        },
                        {
                            "Phoneme": "ɛ",
                            "PronunciationAssessment": {
                                "AccuracyScore": 47.0,
                                "NBestPhonemes": [
                                    {
                                        "Phoneme": "ə",
                                        "Score": 100.0
                                    },
                                    {
                                        "Phoneme": "l",
                                        "Score": 52.0
                                    },
                                    {
                                        "Phoneme": "ɛ",
                                        "Score": 47.0
                                    },
                                    {
                                        "Phoneme": "h",
                                        "Score": 17.0
                                    },
                                    {
                                        "Phoneme": "æ",
                                        "Score": 2.0
                                    }
                                ]
                            },
                            "Offset": 11100000,
                            "Duration": 500000
                        },
                        {
                            "Phoneme": "l",
                            "PronunciationAssessment": {
                                "AccuracyScore": 100.0,
                                "NBestPhonemes": [
                                    {
                                        "Phoneme": "l",
                                        "Score": 100.0
                                    },
                                    {
                                        "Phoneme": "oʊ",
                                        "Score": 46.0
                                    },
                                    {
                                        "Phoneme": "ə",
                                        "Score": 5.0
                                    },
                                    {
                                        "Phoneme": "ɛ",
                                        "Score": 3.0
                                    },
                                    {
                                        "Phoneme": "u",
                                        "Score": 1.0
                                    }
                                ]
                            },
                            "Offset": 11700000,
                            "Duration": 1100000
                        },
                        {
                            "Phoneme": "oʊ",
                            "PronunciationAssessment": {
                                "AccuracyScore": 100.0,
                                "NBestPhonemes": [
                                    {
                                        "Phoneme": "oʊ",
                                        "Score": 100.0
                                    },
                                    {
                                        "Phoneme": "d",
                                        "Score": 29.0
                                    },
                                    {
                                        "Phoneme": "t",
                                        "Score": 24.0
                                    },
                                    {
                                        "Phoneme": "n",
                                        "Score": 22.0
                                    },
                                    {
                                        "Phoneme": "l",
                                        "Score": 18.0
                                    }
                                ]
                            },
                            "Offset": 12900000,
                            "Duration": 8400000
                        }
                    ]
                }
            ]
        }
    ]
}

可获取以下项的发音评估分数：

• 全文
• 单词
• 音节组
• SAPI 或 IPA 格式的音素

每个区域设置支持的功能

下表汇总了区域设置支持的功能。 有关详细信息，请参阅以下部分。 如果下表中未列出支持功能所需的区域设置，请填写此引入表单以获取进一步帮助。

音节组

发音评估可以提供音节级的评估结果。 单词通常是逐个音节发音，而不是逐个音素发音。 按音节分组更清晰，也更符合说话习惯。

发音评估仅通过 IPA 和 SAPI 支持 en-US 中的音节组合。

下表将示例音素与相应的音节进行了比较。

要请求音节级结果和音素，请将粒度配置参数设置为 Phoneme。

音素字母格式

发音评估通过 IPA 支持 en-US 中的音素名称，并通过 SAPI 支持 en-US 和 zh-cn 中的音素名称。

对于支持音素名称的区域设置，音素名称与分数一起提供。 音素名称有助于识别哪些音素发音准确，哪些不准确。 对于其他区域设置，只能获取音素分数。

下表将示例 SAPI 音素与相应的 IPA 音素进行了比较。

要请求 IPA 音素，请将音素字母设置为 IPA。 如果不指定字母表，则默认情况下音素为 SAPI 格式。

pronunciationAssessmentConfig.PhonemeAlphabet = "IPA";

auto pronunciationAssessmentConfig = PronunciationAssessmentConfig::CreateFromJson("{\"referenceText\":\"good morning\",\"gradingSystem\":\"HundredMark\",\"granularity\":\"Phoneme\",\"phonemeAlphabet\":\"IPA\"}");

PronunciationAssessmentConfig pronunciationAssessmentConfig = PronunciationAssessmentConfig.fromJson("{\"referenceText\":\"good morning\",\"gradingSystem\":\"HundredMark\",\"granularity\":\"Phoneme\",\"phonemeAlphabet\":\"IPA\"}");

pronunciation_assessment_config = speechsdk.PronunciationAssessmentConfig(json_string="{\"referenceText\":\"good morning\",\"gradingSystem\":\"HundredMark\",\"granularity\":\"Phoneme\",\"phonemeAlphabet\":\"IPA\"}")

var pronunciationAssessmentConfig = SpeechSDK.PronunciationAssessmentConfig.fromJSON("{\"referenceText\":\"good morning\",\"gradingSystem\":\"HundredMark\",\"granularity\":\"Phoneme\",\"phonemeAlphabet\":\"IPA\"}");

pronunciationAssessmentConfig.phonemeAlphabet = @"IPA";

pronunciationAssessmentConfig?.phonemeAlphabet = "IPA"

评估口语音素

通过口语音素，可获得表明口语音素与预期音素匹配的可能性的置信度分数。

发音评估通过 IPA 和 SAPI 支持 en-US 中的口语音素。

例如，若要获取单词 Hello 的完整口语发音，可以连接每个具有最高置信度分数的预期音素的第一个口语音素。 在以下评估结果中，当你说 hello 这个词时，预期的 IPA 音素是 h ɛ l oʊ。 但是，实际口语音素是 h ə l oʊ。 在本示例中，每个预期的音素都有五个可能的候选项。 评估结果显示最可能的口语音素是 ə 而不是预期的音素 ɛ。 预期的音素 ɛ 仅获得 47 分的置信度。 其他潜在匹配的置信度分数为 52、17 和 2。

{
    "Id": "bbb42ea51bdb46d19a1d685e635fe173",
    "RecognitionStatus": 0,
    "Offset": 7500000,
    "Duration": 13800000,
    "DisplayText": "Hello.",
    "NBest": [
        {
            "Confidence": 0.975003,
            "Lexical": "hello",
            "ITN": "hello",
            "MaskedITN": "hello",
            "Display": "Hello.",
            "PronunciationAssessment": {
                "AccuracyScore": 100,
                "FluencyScore": 100,
                "CompletenessScore": 100,
                "PronScore": 100
            },
            "Words": [
                {
                    "Word": "hello",
                    "Offset": 7500000,
                    "Duration": 13800000,
                    "PronunciationAssessment": {
                        "AccuracyScore": 99.0,
                        "ErrorType": "None"
                    },
                    "Syllables": [
                        {
                            "Syllable": "hɛ",
                            "PronunciationAssessment": {
                                "AccuracyScore": 91.0
                            },
                            "Offset": 7500000,
                            "Duration": 4100000
                        },
                        {
                            "Syllable": "loʊ",
                            "PronunciationAssessment": {
                                "AccuracyScore": 100.0
                            },
                            "Offset": 11700000,
                            "Duration": 9600000
                        }
                    ],
                    "Phonemes": [
                        {
                            "Phoneme": "h",
                            "PronunciationAssessment": {
                                "AccuracyScore": 98.0,
                                "NBestPhonemes": [
                                    {
                                        "Phoneme": "h",
                                        "Score": 100.0
                                    },
                                    {
                                        "Phoneme": "oʊ",
                                        "Score": 52.0
                                    },
                                    {
                                        "Phoneme": "ə",
                                        "Score": 35.0
                                    },
                                    {
                                        "Phoneme": "k",
                                        "Score": 23.0
                                    },
                                    {
                                        "Phoneme": "æ",
                                        "Score": 20.0
                                    }
                                ]
                            },
                            "Offset": 7500000,
                            "Duration": 3500000
                        },
                        {
                            "Phoneme": "ɛ",
                            "PronunciationAssessment": {
                                "AccuracyScore": 47.0,
                                "NBestPhonemes": [
                                    {
                                        "Phoneme": "ə",
                                        "Score": 100.0
                                    },
                                    {
                                        "Phoneme": "l",
                                        "Score": 52.0
                                    },
                                    {
                                        "Phoneme": "ɛ",
                                        "Score": 47.0
                                    },
                                    {
                                        "Phoneme": "h",
                                        "Score": 17.0
                                    },
                                    {
                                        "Phoneme": "æ",
                                        "Score": 2.0
                                    }
                                ]
                            },
                            "Offset": 11100000,
                            "Duration": 500000
                        },
                        {
                            "Phoneme": "l",
                            "PronunciationAssessment": {
                                "AccuracyScore": 100.0,
                                "NBestPhonemes": [
                                    {
                                        "Phoneme": "l",
                                        "Score": 100.0
                                    },
                                    {
                                        "Phoneme": "oʊ",
                                        "Score": 46.0
                                    },
                                    {
                                        "Phoneme": "ə",
                                        "Score": 5.0
                                    },
                                    {
                                        "Phoneme": "ɛ",
                                        "Score": 3.0
                                    },
                                    {
                                        "Phoneme": "u",
                                        "Score": 1.0
                                    }
                                ]
                            },
                            "Offset": 11700000,
                            "Duration": 1100000
                        },
                        {
                            "Phoneme": "oʊ",
                            "PronunciationAssessment": {
                                "AccuracyScore": 100.0,
                                "NBestPhonemes": [
                                    {
                                        "Phoneme": "oʊ",
                                        "Score": 100.0
                                    },
                                    {
                                        "Phoneme": "d",
                                        "Score": 29.0
                                    },
                                    {
                                        "Phoneme": "t",
                                        "Score": 24.0
                                    },
                                    {
                                        "Phoneme": "n",
                                        "Score": 22.0
                                    },
                                    {
                                        "Phoneme": "l",
                                        "Score": 18.0
                                    }
                                ]
                            },
                            "Offset": 12900000,
                            "Duration": 8400000
                        }
                    ]
                }
            ]
        }
    ]
}

要指示是否以及有多少潜在的口语音素可以获得置信度分数，请将 NBestPhonemeCount 参数设置为整​​数值，例如 5。

pronunciationAssessmentConfig.NBestPhonemeCount = 5;

auto pronunciationAssessmentConfig = PronunciationAssessmentConfig::CreateFromJson("{\"referenceText\":\"good morning\",\"gradingSystem\":\"HundredMark\",\"granularity\":\"Phoneme\",\"phonemeAlphabet\":\"IPA\",\"nBestPhonemeCount\":5}");

PronunciationAssessmentConfig pronunciationAssessmentConfig = PronunciationAssessmentConfig.fromJson("{\"referenceText\":\"good morning\",\"gradingSystem\":\"HundredMark\",\"granularity\":\"Phoneme\",\"phonemeAlphabet\":\"IPA\",\"nBestPhonemeCount\":5}");

pronunciation_assessment_config = speechsdk.PronunciationAssessmentConfig(json_string="{\"referenceText\":\"good morning\",\"gradingSystem\":\"HundredMark\",\"granularity\":\"Phoneme\",\"phonemeAlphabet\":\"IPA\",\"nBestPhonemeCount\":5}")

var pronunciationAssessmentConfig = SpeechSDK.PronunciationAssessmentConfig.fromJSON("{\"referenceText\":\"good morning\",\"gradingSystem\":\"HundredMark\",\"granularity\":\"Phoneme\",\"phonemeAlphabet\":\"IPA\",\"nBestPhonemeCount\":5}");

pronunciationAssessmentConfig.nbestPhonemeCount = 5;

pronunciationAssessmentConfig?.nbestPhonemeCount = 5

发音分数计算

发音分数是根据针对阅读和演讲场景的特定公式对准确性、韵律性、流畅性和完整性分数进行加权来计算的。

当将准确性、韵律性、流畅性和完整性的分数从低到高排序（如果每个分数都可用），并将最低分数到最高分数表示为 s0 到 s3 时，发音分数的计算如下：

对于阅读场景：

• 有韵律性分数时：发音分数 = 0.4 * s0 + 0.2 * s1 + 0.2 * s2 + 0.2 * s3
• 无韵律性分数时：发音分数 = 0.6 * s0 + 0.2 * s1 + 0.2 * s2

对于演讲场景（完整性分数不适用）：

• 有韵律性分数时：发音分数 = 0.6 * s0 + 0.2 * s1 + 0.2 * s2
• 无韵律性分数时：发音分数 = 0.6 * s0 + 0.4 * s1

此公式根据每个分数的重要性提供加权计算，确保对发音进行全面评估。

相关内容

• 了解质量基准。
• 尝试在工作室中进行发音评估。
• 查看易于部署的发音评估演示。