Added unconditional generation

app.py

@@ -184,6 +184,7 @@ def infer_from_audio(text, language, accent, audio_prompt, record_audio_prompt,
         audio_prompts = torch.zeros([1, 0, NUM_QUANTIZERS]).type(torch.int32).to(device)
         text_prompts = torch.zeros([1, 0]).type(torch.int32)
         lang_pr = language if language != 'mix' else 'en'
+        enroll_x_lens = 0
     else:
         audio_prompt = audio_prompt if audio_prompt is not None else record_audio_prompt
         sr, wav_pr = audio_prompt