STT for es-MX keeps timing out (westeurope)

I have a webapp that uses 22 languages available on Azure Speech Services.

They all use the same code and all work fine, except for: es-MX. I keep getting timeouts for this language. I don’t know when this started (I suspect this morning.)

Here is as minimal a script I could come up with:

"""
Run speech-to-text on Azure Speech using Python, from the CLI.

python3.11 minitests/azure_stt.py minitests/minitests_data/hola-que-tal.wav es-ES
"""

import argparse
import requests
import base64
import json
import os

import dotenv
dotenv.load_dotenv()



def assess_pronunciation_via_azure_rest(audio_filepath, language, reference_text=None):
    def get_chunk(audio_source, chunk_size=1024):
        while True:
            chunk = audio_source.read(chunk_size)
            if not chunk:
                break
            yield chunk

    audio_file = open(audio_filepath, 'rb') # Remember it won't work in 'wb'
    url=get_azure_rest_assessment_endpoint(language)
    headers=get_azure_rest_assessment_headers(reference_text)

    print('language:', language)
    print('url:', url)
    print('headers:', headers)

    try:
        rest_api_response = requests.post(
            url=url,
            headers=headers,
            data=get_chunk(audio_file),
            timeout=60, # in seconds
        )
    except requests.Timeout:
        return 'timeout'
    except requests.ConnectionError:
        return 'ConnectionError'

    audio_file.close()
    return rest_api_response.json()

def get_azure_rest_assessment_endpoint(language):
    profanity = 'masked' # One of: 'raw', 'masked', 'removed'. But does not do anything. TODO(): report bug.
    region = os.getenv("AZURE_SPEECH_API_REGION")
    endpoint = f"https://{region}.stt.speech.microsoft.com/speech/recognition/conversation/cognitiveservices/v1?language={language}&profanity={profanity}"
    return endpoint

def get_azure_rest_assessment_headers(reference_text):
    api_key = os.getenv("AZURE_SPEECH_API_KEY")
    return {
        'Accept': 'application/json;text/xml',
        'Connection': 'Keep-Alive',
        'Content-Type': 'audio/wav; codecs=audio/pcm; samplerate=16000',
        'Ocp-Apim-Subscription-Key': api_key,
        'Pronunciation-Assessment': get_azure_rest_assessment_parameters(reference_text),
        'Transfer-Encoding': 'chunked',
        'Expect': '100-continue',
    }

def get_azure_rest_assessment_parameters(text):
    parameters = {
        "Dimension": "Comprehensive",
        "EnableMiscue": True,
        "GradingSystem": "HundredMark",
        "Granularity": "Phoneme",
        "nbestPhonemeCount": 3,
        "PhonemeAlphabet": "IPA",
    }
    if text is not None:
        parameters['ReferenceText'] = text

    pronunciation_assessment_parameters_json = json.dumps(parameters) 
    pronunciation_assessment_parameters_base64 = base64.b64encode(bytes(pronunciation_assessment_parameters_json, 'utf-8'))
    pronunciation_assessment_parameters_utf8 = str(pronunciation_assessment_parameters_base64, "utf-8")
    return pronunciation_assessment_parameters_utf8



def main():
    parser = argparse.ArgumentParser(description="Assess pronunciation using Azure REST API")
    parser.add_argument("audio_filepath", help="Path to the audio file")
    parser.add_argument("language", help="Language of the pronunciation")
    parser.add_argument("--reference_text", help="Reference text (optional)", default=None)

    args = parser.parse_args()

    result = assess_pronunciation_via_azure_rest(args.audio_filepath, args.language, args.reference_text)
    print("Assessment Result:", result)

if __name__ == "__main__":
    main()

I use AZURE_SPEECH_API_REGION=westeurope.

Again, works fine for me for various languages… But keeps timing out for es-MX.

Can someone please have a look?

This is urgent & important as it breaks any STT for es-MX and the app is in production.

Thank you!

[chschrae]

I changed the pronunciation assessment code to check for none:

def get_azure_rest_assessment_headers(reference_text):
    api_key = os.getenv("AZURE_SPEECH_API_KEY")
    headers = {
        'Accept': 'application/json;text/xml',
        'Connection': 'Keep-Alive',
        'Content-Type': 'audio/wav; codecs=audio/pcm; samplerate=16000',
        'Ocp-Apim-Subscription-Key': api_key,
        # 'Pronunciation-Assessment': get_azure_rest_assessment_parameters(reference_text),
        'Transfer-Encoding': 'chunked',
        'Expect': '100-continue',
    }
    if reference_text is not None:
        headers['Pronunciation-Assessment'] = get_azure_rest_assessment_parameters(reference_text)
    # return {
    #     'Accept': 'application/json;text/xml',
    #     'Connection': 'Keep-Alive',
    #     'Content-Type': 'audio/wav; codecs=audio/pcm; samplerate=16000',
    #     'Ocp-Apim-Subscription-Key': api_key,
    #     'Pronunciation-Assessment': get_azure_rest_assessment_parameters(reference_text),
    #     'Transfer-Encoding': 'chunked',
    #     'Expect': '100-continue',
    # }
    return headers

It seems to work now. This could be a bug in the service and I would expect the same behavior for all languages. I will contact the service team and file a bug.

I’m going to close this issue for now since it seems there is a work around. If this is still an issue for you, feel free to re-open it.

[wangkenpu]

@fabswt The issue has been fixed. You can retry your original code.