diff --git a/lib/synth-audio.js b/lib/synth-audio.js
index 1e23135..88c992c 100644
--- a/lib/synth-audio.js
+++ b/lib/synth-audio.js
@@ -318,23 +318,24 @@ const synthMicrosoft = async(logger, {
     let speechConfig;
     if (use_custom_tts && custom_tts_endpoint_url) {
       if (apiKey) {
-        speechConfig = SpeechConfig.fromEndpoint(custom_tts_endpoint_url, apiKey);
+        speechConfig = SpeechConfig.fromEndpoint(new URL(custom_tts_endpoint_url), apiKey);
       } else {
-        speechConfig = SpeechConfig.fromEndpoint(custom_tts_endpoint_url);
+        speechConfig = SpeechConfig.fromEndpoint(new URL(custom_tts_endpoint_url));
       }
     } else {
       speechConfig = SpeechConfig.fromSubscription(apiKey, region);
     }
     speechConfig.speechSynthesisLanguage = language;
     speechConfig.speechSynthesisVoiceName = voice;
-    if (use_custom_tts && custom_tts_endpoint) {
+    if (use_custom_tts && custom_tts_endpoint ) {
       speechConfig.endpointId = custom_tts_endpoint;
-
+    }
+    if (use_custom_tts && !content.startsWith('<speak')) {
       /**
        * Note: it seems that to use custom voice ssml is required with the voice attribute
        * Otherwise sending plain text we get "Voice does not match"
        */
-      if (!content.startsWith('<speak')) content = `<speak>${text}</speak>`;
+      content = `<speak>${text}</speak>`;
     }
     speechConfig.speechSynthesisOutputFormat = trimSilence ?
       SpeechSynthesisOutputFormat.Raw8Khz16BitMonoPcm :