local Edge TTS

app/models.py

@@ -52,6 +52,7 @@ AVAILABLE_MODELS = {
 
     # # Microsoft Edge TTS
     # 'innoai/Edge-TTS-Text-to-Speech': 'innoai/Edge-TTS-Text-to-Speech', # API disabled
+    'innoai/Edge-TTS-Text-to-Speech': '/Edge-TTS', # using Edge API
 
     # IMS-Toucan
     # 'Flux9665/MassivelyMultilingualTTS': 'Flux9665/MassivelyMultilingualTTS', # 5.1
@@ -310,6 +311,7 @@ HF_SPACES = {
         'return_audio_index': 0,
         'is_zero_gpu_space': False,
         'series': 'Kokoro',
+        'space_link': 'Remsky/Kokoro-TTS-Zero', # still supports v0.19
     },
 
     # StyleTTS Kokoro v0.23

app/synth.py

@@ -80,7 +80,12 @@ def synthandreturn(text, autoplay, request: gr.Request):
         while attempt_count < max_attempts:
             try:
                 if model in AVAILABLE_MODELS:
-                    if '/' in model:
+                    if '/' == AVAILABLE_MODELS[model][0]:
+                        # local model
+                        # just Edge TTS API
+                        from .tts.edge import edge_text_to_speech
+                        result = edge_text_to_speech(text, 'en-US-EmmaMultilingualNeural - en-US (Female)')
+                    elif '/' in model:
                         # Use public HF Space
                         # if (model not in hf_clients):
                         #     #save client to local variable; can timeout
@@ -169,6 +174,7 @@ def synthandreturn(text, autoplay, request: gr.Request):
         else:
             print('Done with', model)
 
+        # Resample to 24kHz
         try:
             with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as f:
                 audio = AudioSegment.from_file(result)

requirements.txt

@@ -6,3 +6,4 @@ git+https://github.com/unitaryai/detoxify
 pyloudnorm
 langdetect
 pydub
+edge_tts==6.1.12