Troubleshoot file writing problem.

Work with coqui synthesis in memory so that file is never written. Switch espeakng to synthesize on command line with subprocess package.

src/synthesize.py

@@ -8,6 +8,7 @@ import subprocess
 from scipy.io import wavfile
 from transformers import pipeline
 import os
+import numpy as np
 
 def synth_mms(text:str, model:str):
     '''
@@ -42,7 +43,9 @@ def synth_coqui(text:str, model:str):
         text: Text to synthesze
         model: Model code 
     Returns:
-        Streaming Wav and sampling rate.
+        Streaming Wav and sampling rate. 
+        
+    IMPORTANT: Current implementation assumes 22050 sampling rate, this should be verified when adding a new model.
     '''
     if model is not None:
         # Get device
@@ -50,14 +53,11 @@ def synth_coqui(text:str, model:str):
         
         # Init TTS
         tts = TTS(model, progress_bar=False).to(device)
-    
-        tts.tts_to_file(text=text, file_path="test.wav", is_multi_speaker=False)
-        
-        sampling_rate, wav = wavfile.read('test.wav')
-        os.remove("test.wav")
+
+        # Infer
+        wav = tts.tts(text=text) # is_multi_speaker=False
         
-        #wav = tts.tts(text=text)
-        return wav, sampling_rate
+        return np.array(wav), 22050
     else:
         return None
 
@@ -74,12 +74,11 @@ def synth_espeakng(text:str, model:str):
     '''
     if model is not None:
         
-        #subprocess.run(['espeak-ng', f'-v{model}', "-w test.wav", text]) #.returncode
-        esng = espeakng.Speaker()
-        esng.voice = model
-        esng.say(text, export_path="test.wav")
+        subprocess.run(['espeak-ng', f'-v{model}', "-w test.wav", text]) 
+        #esng = espeakng.Speaker()
+        #esng.voice = model
+        #esng.say(text, export_path="test.wav")
 
-        print(os.listdir())
         sampling_rate, wav = wavfile.read('test.wav')
         os.remove("test.wav")