Update app.py

app.py

@@ -3,16 +3,26 @@ import gradio as gr
 from datetime import datetime
 import pytz
 import os
-import elevenlabs
+import numpy
+from elevenlabs import voices, generate, set_api_key, UnauthenticatedRateLimitError
 
-elevenlabsKey = os.environ.get("ELEVEN_KEY_TOKEN", None)
-voice = os.environ.get("VOICE_ID", None)
+
+elevenlabs_api_key = os.environ.get("ELEVEN_KEY_TOKEN", None)
+voice_id = os.environ.get("VOICE_ID", None)
 model = "eleven_multilingual_v2"
 ttsPassword = os.environ.get("TTS_PASSWORD", None)
 enable_tts = False
 
 client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
 
+def pad_buffer(audio):
+    # Pad buffer to multiple of 2 bytes
+    buffer_size = len(audio)
+    element_size = np.dtype(np.int16).itemsize
+    if buffer_size % element_size != 0:
+        audio = audio + b'\0' * (element_size - (buffer_size % element_size))
+    return audio 
+
 def format_prompt(message, history, system_prompt=None):
     prompt = "<s>"
     
@@ -27,6 +37,19 @@ def format_prompt(message, history, system_prompt=None):
     prompt += f"[INST] {message} [/INST]"
     return prompt
 
+def generate_voice(text, voice_name, api_key):
+    set_api_key(elevenlabs_api_key) #set API key
+    try:
+        audio = generate(
+            voice=voice_name, 
+            model="eleven_multilingual_v2"
+        )
+        return (44100, np.frombuffer(pad_buffer(audio), dtype=np.int16))
+    except UnauthenticatedRateLimitError as e:
+        raise gr.Error("Thanks for trying out ElevenLabs TTS! You've reached the free tier limit. Please provide an API key to continue.") 
+    except Exception as e:
+        raise gr.Error(e)
+
 def generate(
     prompt, history, enable_tts, tts_password, temperature=0.2, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0
 ):
@@ -115,18 +138,8 @@ def generate(
     return output
 
         # Check if the checkbox is selected and the correct password is provided
-    if enable_tts and passwordInput == ttsPassword:
-        # Generate the audio file
-        audio = elevenlabs.generate(response, voice, model, api_key=elevenlabsKey)
-            
-        # Play the audio file
-        elevenlabs.play(audio)
-
-    # Return the audio file as a numpy array
-        return audio
-    else:
-        # Return None if the TTS is not enabled or the password is incorrect
-        return None
+if enable_tts and passwordInput == ttsPassword:
+    generate_voice(output, voice_id)
 
 checkbox = gr.Checkbox(label="Enable TTS")
 passwordInput = gr.Textbox(label="TTS Password", type="password")