:clown_face: please work

groq-voicechat-demo.py

@@ -2,7 +2,7 @@ import gradio as gr
 from groq import Groq
 import os
 from whisper_support import transcribe
-from kokoro_support import generate_tts
+from kokoro_support import generate_tts, voice_choices
 
 groq_api_key = os.environ.get("GROQ_API_KEY")
 
@@ -129,6 +129,18 @@ def create_demo():
 
         with gr.Row():
             playback_button = gr.Button("playback last message")
+        with gr.Row():
+            voice_opps = gr.Dropdown(
+                choices = list(voice_choices.keys()),
+                value = list(voice_choices.keys())[0]
+            )
+        with gr.Row():
+            voice_speed = gr.Slider(
+                minimum = 0.5,
+                maximum = 2,
+                value = 1,
+                step = 0.1,
+            )
         with gr.Row():
             audio_out = gr.Audio(
                 label = "Output Audio",
@@ -136,22 +148,6 @@ def create_demo():
                 autoplay = True,
                 streaming = True
             )
-
-
-        
-        # stream = voiceinput.start_recording(
-        #     process_audio,
-        #     [voiceinput],
-        #     [voiceinput],
-        # )
-        # respond = voiceinput.stop_recording(
-        #     groq_voicechat, 
-        #     [voiceinput, chatbot, model, system_prompt,  ],
-        #     [chatbot]
-        # )
-
-
-
         
         output = voiceinput.stop_recording(
             groq_voicechat, 
@@ -159,10 +155,10 @@ def create_demo():
             [chatbot]
         ) #WHAT AM I DOING LOL - COME BACK TO THIS
 
-        def playback_last_message(chat_history):
+        def playback_last_message(chat_history, voice, speed):
             if len(chat_history) > 0:
                 last_message = chat_history[-1]['content'][0]['text']
-                gen_object = generate_tts(last_message)
+                gen_object = generate_tts(last_message, voice_choices[voice], speed)
                 for chunk in gen_object:
                     yield chunk
             return None
@@ -171,7 +167,7 @@ def create_demo():
         
         playback_button.click(
             playback_last_message,
-            inputs=[chatbot],
+            inputs=[chatbot, voice_opps, voice_speed],
             outputs=[audio_out]
         )
 

kokoro_support.py

@@ -5,30 +5,46 @@ import numpy as np
 import kokoro
 import torch
 
+voice_choices = {
+    '🇺🇸 🚺 Heart ❤️': 'af_heart',
+    '🇺🇸 🚺 Bella 🔥': 'af_bella',
+    '🇺🇸 🚺 Nicole 🎧': 'af_nicole',
+    '🇺🇸 🚺 Aoede': 'af_aoede',
+    '🇺🇸 🚺 Kore': 'af_kore',
+    '🇺🇸 🚺 Sarah': 'af_sarah',
+    '🇺🇸 🚺 Nova': 'af_nova',
+    '🇺🇸 🚺 Sky': 'af_sky',
+    '🇺🇸 🚺 Alloy': 'af_alloy',
+    '🇺🇸 🚺 Jessica': 'af_jessica',
+    '🇺🇸 🚺 River': 'af_river',
+    '🇺🇸 🚹 Michael': 'am_michael',
+    '🇺🇸 🚹 Fenrir': 'am_fenrir',
+    '🇺🇸 🚹 Puck': 'am_puck',
+    '🇺🇸 🚹 Echo': 'am_echo',
+    '🇺🇸 🚹 Eric': 'am_eric',
+    '🇺🇸 🚹 Liam': 'am_liam',
+    '🇺🇸 🚹 Onyx': 'am_onyx',
+    '🇺🇸 🚹 Santa': 'am_santa',
+    '🇺🇸 🚹 Adam': 'am_adam',
+    '🇬🇧 🚺 Emma': 'bf_emma',
+    '🇬🇧 🚺 Isabella': 'bf_isabella',
+    '🇬🇧 🚺 Alice': 'bf_alice',
+    '🇬🇧 🚺 Lily': 'bf_lily',
+    '🇬🇧 🚹 George': 'bm_george',
+    '🇬🇧 🚹 Fable': 'bm_fable',
+    '🇬🇧 🚹 Lewis': 'bm_lewis',
+    '🇬🇧 🚹 Daniel': 'bm_daniel',
+    }
 
 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([])
 
+pipelines = {lang_code: KPipeline(lang_code=lang_code, model=False) for lang_code in 'ab'}
+pipelines['a'].g2p.lexicon.golds['kokoro'] = 'kˈOkəɹO'
+pipelines['b'].g2p.lexicon.golds['kokoro'] = 'kˈQkəɹQ'
 
 def generate_tts(text, voice='af_heart', speed=1):
+    pipeline = pipelines[voice[0]]
     pack = pipeline.load_voice(voice)
     first = True
     for _, ps, _ in pipeline(text, voice, speed):