Pronunciation Assessment SDK returns low score with ErrorType: 'Mispronunciation' compared to Speech Studio.

Question

In my NodeJS app and microsoft-cognitiveservices-speech-sdk package, Azure's Pronunciation Assessment for Japanese (It works fine when I test it in English.) always returns ErrorType: 'Mispronunciation' and a very low score (around 50) even though it usually returns around 90-100 when tested on Speech Studio.

I tried making the audio format to be the same as what's used on Studio.
channels = 1, bitsPerSample = 16, samplesPerSecond = 16000
I also tried downloading the audio file from the studio and tested it with my app which also returns the same result.

{
    Word: 'こんにちは',
    Offset: 5700000,
    Duration: 11700000,
    PronunciationAssessment: { AccuracyScore: 53, ErrorType: 'Mispronunciation' },
    Phonemes: [
      [Object], [Object],
      [Object], [Object],
      [Object], [Object],
      [Object], [Object],
      [Object]
    ]
  }

{   
 accuracyScore: 54,  
 fluencyScore: 100,   
completenessScore: 0,   
pronunciationScore: 30.8 
}

completenessScore is always 0 as well.

Here is my code.

async azureGradeSpeech(
    input: GradeSpeechInput
  ): Promise<SpeechScore> {
    const { text, audioFilePath } = input;
    const speechConfig = SpeechSDK.SpeechConfig.fromSubscription(
      azureSubscriptionKey,
      azureRegion
    );

    const buffer = await this.readLocalFileAsBuffer(audioFilePath);

    const audioConfig = SpeechSDK.AudioConfig.fromWavFileInput(buffer);
    const audAudioConfig =
      SpeechSDK.AutoDetectSourceLanguageConfig.fromLanguages(["ja-JP"]);

    const speechRecognizer = SpeechSDK.SpeechRecognizer.FromConfig(
      speechConfig,
      audAudioConfig,
      audioConfig
    );

    const resultConfig = {
      referenceText: text,
      gradingSystem: "HundredMark",
      granularity: "Phoneme", // Phoneme, Syllable, Word, FullText
      // EnableMiscue: true,
    };

    const pronunciationAssessmentConfig =
      SpeechSDK.PronunciationAssessmentConfig.fromJSON(
        JSON.stringify(resultConfig)
      );

    pronunciationAssessmentConfig.applyTo(speechRecognizer);

    const result: SpeechScore = await new Promise((resolve, reject) => {
      speechRecognizer.recognizeOnceAsync(
        (speechRecognitionResult: SpeechSDK.SpeechRecognitionResult) => {
          // The pronunciation assessment result as a Speech SDK object
          const pronunciationAssessmentResult =
            SpeechSDK.PronunciationAssessmentResult.fromResult(
              speechRecognitionResult
            );

          const pronunciationAssessmentResultJson =
            speechRecognitionResult.properties.getProperty(
              SpeechSDK.PropertyId.SpeechServiceResponse_JsonResult
            );

          const jsonResult = JSON.parse(pronunciationAssessmentResultJson);

          const words = jsonResult.NBest[0].Words;
          const length = words.length;
          const totalAccuracyScore = words.reduce(
            (accumulator, curr) =>
              curr.PronunciationAssessment.AccuracyScore + accumulator,
            0
          );
          const average = Number((totalAccuracyScore / length).toFixed(0));

          const score: SpeechScore = {
            averageAccuracyScore: average,
            accuracyScore: pronunciationAssessmentResult.accuracyScore,
            fluencyScore: pronunciationAssessmentResult.fluencyScore,
            completenessScore: pronunciationAssessmentResult.completenessScore,
            pronunciationScore:
              pronunciationAssessmentResult.pronunciationScore,
          };

          resolve(score);
        },
        (error) => {
          reject(error);
        }
      );
    });

    console.log("result", result);
    return result;
  }