Azure语音转文本西班牙语识别异常：设置es-ES仍识别为英文求助

我用Python编写了如下Azure Function，已将语音识别语言设置为es-ES，但测试时系统仍将西班牙语语音识别为英文——比如我说“Soy Anna”，识别结果为“Sorry, Anna”。不确定是设置有误还是按需付费（Pay-As-You-Go）层不支持西班牙语，多次调整设置未解决，希望实现西班牙语语音的正确识别。

代码如下：

import logging
import azure.functions as func
import azure.cognitiveservices.speech as speechsdk
import azure_config

def main(req: func.HttpRequest) -> func.HttpResponse:

    language = 'es-ES' 

    speech_config = speechsdk.SpeechConfig(subscription=azure_config.speech_subscription_key, region=azure_config.speech_region, speech_recognition_language=language)

    audio_config = speechsdk.audio.AudioConfig(use_default_microphone=True)
    speech_recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, audio_config=audio_config)

    logging.info("Speak into your microphone.")
    speech_recognition_result = speech_recognizer.recognize_once_async().get()

    if speech_recognition_result.reason == speechsdk.ResultReason.RecognizedSpeech:
        logging.info("Recognized: {}".format(speech_recognition_result.text))

    elif speech_recognition_result.reason == speechsdk.ResultReason.NoMatch:
        logging.info("No speech could be recognized: {}".format(speech_recognition_result.no_match_details))

    elif speech_recognition_result.reason == speechsdk.ResultReason.Canceled:
        cancellation_details = speech_recognition_result.cancellation_details
        logging.info("Speech Recognition canceled: {}".format(cancellation_details.reason))
        
        if cancellation_details.reason == speechsdk.CancellationReason.Error:
            logging.info("Error details: {}".format(cancellation_details.error_details))
            logging.info("Did you set the speech resource key and region values?")
    
    return func.HttpResponse("Speech recognition completed.", status_code=200)

排查与解决方案 #

1. 核心问题：Azure Function无法访问本地麦克风 #

你的代码中使用了use_default_microphone=True，这在本地调试时可以调用你电脑的麦克风，但部署到Azure云端后，Function运行在服务器环境中，没有可用的麦克风硬件，实际接收的是服务器环境的噪音，因此被误识别为英文内容。

修正方案：修改Function接收客户端上传的音频数据（比如音频文件或流），再进行语音识别，示例代码如下：

import logging
import azure.functions as func
import azure.cognitiveservices.speech as speechsdk
import azure_config

def main(req: func.HttpRequest) -> func.HttpResponse:
    language = 'es-ES' 
    speech_config = speechsdk.SpeechConfig(subscription=azure_config.speech_subscription_key, region=azure_config.speech_region, speech_recognition_language=language)

    # 从HTTP请求中获取音频数据
    audio_data = req.get_body()
    if not audio_data:
        return func.HttpResponse("未提供音频数据", status_code=400)
    
    # 创建推送音频流并写入数据
    push_stream = speechsdk.audio.PushAudioInputStream()
    push_stream.write(audio_data)
    audio_config = speechsdk.audio.AudioConfig(stream=push_stream)
    
    speech_recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, audio_config=audio_config)

    logging.info("开始处理音频识别")
    speech_recognition_result = speech_recognizer.recognize_once_async().get()

    # 结果处理逻辑保持不变
    if speech_recognition_result.reason == speechsdk.ResultReason.RecognizedSpeech:
        recognized_text = speech_recognition_result.text
        logging.info(f"识别结果: {recognized_text}")
        return func.HttpResponse(f"识别结果：{recognized_text}", status_code=200)

    elif speech_recognition_result.reason == speechsdk.ResultReason.NoMatch:
        msg = f"无法识别语音: {speech_recognition_result.no_match_details}"
        logging.info(msg)
        return func.HttpResponse(msg, status_code=400)

    elif speech_recognition_result.reason == speechsdk.ResultReason.Canceled:
        cancellation_details = speech_recognition_result.cancellation_details
        msg = f"语音识别被取消: {cancellation_details.reason}"
        logging.info(msg)
        if cancellation_details.reason == speechsdk.CancellationReason.Error:
            msg += f"\n错误详情: {cancellation_details.error_details}"
            logging.info(msg)
        return func.HttpResponse(msg, status_code=500)

2. 验证语言设置是否生效 #

可以在创建SpeechConfig后添加日志，确认语言参数是否正确加载：

logging.info(f"当前识别语言: {speech_config.speech_recognition_language}")

如果输出不是es-ES，检查azure_config模块是否有覆盖设置的逻辑。

3. 确认区域与语言兼容性 #

Pay-As-You-Go层完全支持西班牙语语音识别，只需确认你的语音资源所在区域（speech_region）支持西班牙语。可以在Azure门户的语音资源页面查看该区域支持的语言列表，确保es-ES在列。

4. 确保音频质量 #

客户端上传的音频需要满足：

• 采样率16kHz、16位单声道PCM格式（推荐）
• 无明显杂音，发音清晰标准

内容的提问来源于stack exchange，提问作者greyhound131