:clown_face: streaming commit

README.md

@@ -1,2 +1,28 @@
-# voicechat-demo
-quick voicechat demo with manual options for playback and input
+# ---
+
+# title: Groq Voicechat Demo
+
+# emoji: 💻
+
+# colorFrom: indigo
+
+# colorTo: pink
+
+# sdk: gradio
+
+# sdk\_version: 6.0.2
+
+# app\_file: groq-voicechat-demo.py
+
+# pinned: false
+
+# license: gpl-3.0
+
+# short\_description: llm backed voicechat with kokoro and whisper
+
+# ---
+
+# 
+
+# Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+

groq-voicechat-demo.py

@@ -133,7 +133,8 @@ def create_demo():
             audio_out = gr.Audio(
                 label = "Output Audio",
                 interactive = False,
-                autoplay = True
+                autoplay = True,
+                streaming = True
             )
 
 
@@ -161,8 +162,12 @@ def create_demo():
         def playback_last_message(chat_history):
             if len(chat_history) > 0:
                 last_message = chat_history[-1]['content'][0]['text']
-                return generate_tts(last_message)
+                gen_object = generate_tts(last_message)
+                for chunk in gen_object:
+                    yield chunk
             return None
+            #     return generate_tts(last_message)
+            # return None
         
         playback_button.click(
             playback_last_message,

kokoro_support.py

@@ -3,26 +3,42 @@ import os
 import random
 import numpy as np
 import kokoro
+import torch
 
 
 model = KModel().to('cpu').eval()
 pipeline = KPipeline(lang_code='a', model=False)
     
+# def generate_tts(text, voice='af_heart', speed=1):
+#     pack = pipeline.load_voice(voice)
+#     audio_chunks = []
+#     for _, ps, _ in pipeline(text, voice, speed):
+#         ref_s = pack[len(ps)-1]
+#         try:
+#             audio = model(ps, ref_s, speed)
+#             audio_chunks.append(audio.numpy())
+#         except:
+#             print("lol there was an issue idk")
+#         # yield 24000, audio.numpy()
+#     if audio_chunks:
+#         concatenated_audio = np.concatenate(audio_chunks)
+#         print(concatenated_audio.shape)
+#         return 24000, concatenated_audio
+#     else:
+#         return 24000, np.array([])
+
+
 def generate_tts(text, voice='af_heart', speed=1):
     pack = pipeline.load_voice(voice)
-    audio_chunks = []
+    first = True
     for _, ps, _ in pipeline(text, voice, speed):
         ref_s = pack[len(ps)-1]
         try:
             audio = model(ps, ref_s, speed)
-            audio_chunks.append(audio.numpy())
         except:
             print("lol there was an issue idk")
-        # yield 24000, audio.numpy()
-    if audio_chunks:
-        concatenated_audio = np.concatenate(audio_chunks)
-        print(concatenated_audio.shape)
-        return 24000, concatenated_audio
-    else:
-        return 24000, np.array([])
+        yield 24000, audio.numpy()
+        if first:
+            first = False
+            yield 24000, torch.zeros(1).numpy()