Back to 2x :)

app.py

@@ -124,7 +124,7 @@ def greet(Text,Voicetoclone,VoiceMicrophone):
     else:
       os.system('ffmpeg-normalize $sample -nt rms -t=-27 -o $sample -ar 16000 -f')
       reference_emb = SE_speaker_manager.compute_d_vector_from_clip(reference_files)
-      model.length_scale = 3  # scaler for the duration predictor. The larger it is, the slower the speech.
+      model.length_scale = 2  # scaler for the duration predictor. The larger it is, the slower the speech.
       model.inference_noise_scale = 0.3 # defines the noise variance applied to the random z vector at inference.
       model.inference_noise_scale_dp = 0.3 # defines the noise variance applied to the duration predictor z vector at inference.
       text = text