CODEWITHSUNDEEP
text-to-speechwireshark
tzviya
tzviya

Reputation: 568

Inconsistent HTTP Visibility in Wireshark for Azure API Calls from C# vs. Python and CURL

I am using the Azure API to get audio for words. This is my code in C#, and it worked fine until I noticed that for some words it returns empty audio.

       string subscriptionKey = Environment.GetEnvironmentVariable("STT_API_KEY");
       Authentication auth = new Authentication("https://eastus.api.cognitive.microsoft.com/sts/v1.0/issueToken", subscriptionKey);
       try
       {
           accessToken = await auth.FetchTokenAsync().ConfigureAwait(false);
           //Console.WriteLine("Successfully obtained an access token. \n");
       }
       catch (Exception ex)
       {
           //Console.WriteLine("Failed to obtain an access token.");
           //Console.WriteLine(ex.ToString());
           //Console.WriteLine(ex.Message);
           return null;
       }

       string host = "https://eastus.tts.speech.microsoft.com/cognitiveservices/v1";

       // Create SSML document.
       XDocument body = new XDocument(
               new XElement("speak",
                   new XAttribute("version", "1.0"),
                   new XAttribute(XNamespace.Xml + "lang", "en-US"),
                   new XElement("voice",
                       new XAttribute(XNamespace.Xml + "lang", "en-US"),
                       new XAttribute(XNamespace.Xml + "gender", "Female"),
                       new XAttribute("name", "en-US-AvaNeural"), // Short name for "Microsoft Server Speech Text to Speech Voice (en-US, Jessa24KRUS)"
                       text)));

       using (HttpClient client = new HttpClient())
       {
           using (HttpRequestMessage request = new HttpRequestMessage())
           {
               // Set the HTTP method
               request.Method = HttpMethod.Post;
               // Construct the URI
               request.RequestUri = new Uri(host);
               // Set the content type header
               request.Content = new StringContent(body.ToString(), Encoding.UTF8, "application/ssml+xml");
               // Set additional header, such as Authorization and User-Agent
               request.Headers.Add("Authorization", "Bearer " + accessToken);
               request.Headers.Add("Connection", "Keep-Alive");
               // Update your resource name
               request.Headers.Add("User-Agent", "YOUR_RESOURCE_NAME");
               // Audio output format. See API reference for full list.
               request.Headers.Add("X-Microsoft-OutputFormat", "riff-24khz-16bit-mono-pcm");
               // Create a request
               //Console.WriteLine("Calling the TTS service. Please wait... \n");
               using (HttpResponseMessage response = await client.SendAsync(request).ConfigureAwait(false))

To investigate the issue, I tried calling the same API through other means: directly via CURL and also through Python with this code.

endpoint = 'https://eastus.tts.speech.microsoft.com/cognitiveservices/v1'
ssml = """
<speak version='1.0' xml:lang='en-US'>
  <voice xml:lang='en-US' xml:gender='Female' name='en-US-AvaNeural'>
    black
  </voice>
</speak>
"""

headers = {
    'Ocp-Apim-Subscription-Key': subscription_key,
    'Content-Type': 'application/ssml+xml',
    'X-Microsoft-OutputFormat': 'audio-16khz-32kbitrate-mono-mp3'
}

response = requests.post(endpoint, headers=headers, data=ssml, verify=False)

To my surprise, I was able to get the audio for words that I couldn't get through C#.

I wanted to compare the requests to understand why I am not getting the same results in different cases. To do this, I opened Wireshark and configured TLS to use SSLKEYLOGFILE so that I could see the frames unencrypted. I ran the request through Python and saw in Wireshark that an HTTP request with all the parameters I set was being made. However, when I ran the request through C# and filtered by the address, I didn't see any HTTP request at all (only TCP and TLS), and the same happened with CURL.

I tried to find the reason why HTTP requests might not be visible, and the answers I saw suggested it was because the traffic is encrypted. However, I saw that the decryption worked when the code was run from Python. Why should there be any difference based on which code the request originates from?

Upvotes: 1

Views: 55

Answers (1)

Josef
Josef

Reputation: 1532

There is currently no support in the .NET Framework for writing the pre-master secret to a file that Wireshark needs to decrypt TLS.

There are a few options:

  1. use an intercepting proxy such as mitmproxy to decrypt the traffic
  2. write the decrypted data from your C# application using PcapStream and open that in Wireshark.

Maybe the discussions in this question or this github topic can help you.

Upvotes: 1

Related Questions