Update app.py
Browse files
app.py
CHANGED
@@ -11,8 +11,6 @@ import random
|
|
11 |
from openai import OpenAI
|
12 |
import subprocess
|
13 |
from starlette.requests import ClientDisconnect
|
14 |
-
import logging
|
15 |
-
import time
|
16 |
|
17 |
LLAMA_3B_API_ENDPOINT = os.environ.get("LLAMA_3B_API_ENDPOINT")
|
18 |
LLAMA_3B_API_KEY = os.environ.get("LLAMA_3B_API_KEY")
|
@@ -20,8 +18,6 @@ HF_TOKEN = os.environ.get("HF_TOKEN", None)
|
|
20 |
|
21 |
default_lang = "en"
|
22 |
engines = { default_lang: Model(default_lang) }
|
23 |
-
logging.basicConfig(level=logging.INFO)
|
24 |
-
logger = logging.getLogger(__name__)
|
25 |
|
26 |
LANGUAGE_CODES = {
|
27 |
"English": "eng",
|
@@ -124,76 +120,64 @@ def models(text, model="Llama 3 8B Service", seed=42):
|
|
124 |
|
125 |
return output
|
126 |
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
output_file = f"translated_audio_{int(time.time())}.wav"
|
131 |
-
|
132 |
-
command = [
|
133 |
-
"expressivity_predict",
|
134 |
-
audio_file,
|
135 |
-
"--tgt_lang", language_code,
|
136 |
-
"--model_name", "seamless_expressivity",
|
137 |
-
"--vocoder_name", "vocoder_pretssel",
|
138 |
-
"--gated-model-dir", "models",
|
139 |
-
"--output_path", output_file
|
140 |
-
]
|
141 |
|
142 |
-
|
143 |
-
|
144 |
-
stdout=asyncio.subprocess.PIPE,
|
145 |
-
stderr=asyncio.subprocess.PIPE
|
146 |
-
)
|
147 |
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
|
|
|
|
|
|
|
|
153 |
|
154 |
-
|
155 |
-
raise Exception(f"Translation process failed: {stderr.decode()}")
|
156 |
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
except Exception as e:
|
163 |
-
print(f"Translation error: {str(e)}")
|
164 |
return None
|
165 |
|
166 |
async def respond(audio, model, seed, target_language):
|
167 |
try:
|
168 |
if audio is None:
|
169 |
-
return None, None
|
170 |
-
|
171 |
user_input = transcribe(audio)
|
172 |
if not user_input:
|
173 |
-
return None, None
|
174 |
-
|
175 |
if user_input.lower().startswith("please translate"):
|
176 |
-
#
|
177 |
-
|
178 |
-
|
|
|
179 |
else:
|
180 |
reply = models(user_input, model, seed)
|
181 |
communicate = edge_tts.Communicate(reply, voice="en-US-ChristopherNeural")
|
182 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
|
183 |
tmp_path = tmp_file.name
|
184 |
await communicate.save(tmp_path)
|
185 |
-
return tmp_path, None
|
186 |
except ClientDisconnect:
|
187 |
print("Client disconnected")
|
188 |
-
return None, None
|
189 |
except Exception as e:
|
190 |
print(f"An error occurred: {str(e)}")
|
191 |
-
return None, None
|
192 |
|
193 |
def clear_history():
|
194 |
global conversation_history
|
195 |
conversation_history = []
|
196 |
-
return None, None
|
197 |
|
198 |
with gr.Blocks(css="style.css") as demo:
|
199 |
gr.Markdown("# <center><b>Optimus Prime: Voice Assistant with Translation</b></center>")
|
@@ -201,7 +185,6 @@ with gr.Blocks(css="style.css") as demo:
|
|
201 |
|
202 |
with gr.Row():
|
203 |
with gr.Column(scale=1):
|
204 |
-
input_audio = gr.Audio(label="User Input", sources=["microphone"], type="filepath")
|
205 |
select = gr.Dropdown([
|
206 |
'Llama 3 8B Service',
|
207 |
'Mixtral 8x7B',
|
@@ -212,11 +195,6 @@ with gr.Blocks(css="style.css") as demo:
|
|
212 |
value="Llama 3 8B Service",
|
213 |
label="Model"
|
214 |
)
|
215 |
-
target_lang = gr.Dropdown(
|
216 |
-
choices=list(LANGUAGE_CODES.keys()),
|
217 |
-
value="German",
|
218 |
-
label="Target Language for Translation"
|
219 |
-
)
|
220 |
seed = gr.Slider(
|
221 |
label="Seed",
|
222 |
minimum=0,
|
@@ -225,19 +203,26 @@ with gr.Blocks(css="style.css") as demo:
|
|
225 |
value=0,
|
226 |
visible=False
|
227 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
228 |
clear_button = gr.Button("Clear Conversation History")
|
229 |
|
230 |
with gr.Column(scale=1):
|
231 |
output_audio = gr.Audio(label="AI Response", type="filepath", interactive=False, autoplay=True)
|
232 |
translated_audio = gr.Audio(label="Translated Audio", type="filepath", interactive=False, autoplay=True)
|
|
|
233 |
|
234 |
input_audio.change(
|
235 |
fn=respond,
|
236 |
inputs=[input_audio, select, seed, target_lang],
|
237 |
-
outputs=[output_audio, translated_audio],
|
238 |
)
|
239 |
|
240 |
-
clear_button.click(fn=clear_history, inputs=[], outputs=[output_audio, translated_audio])
|
241 |
|
242 |
if __name__ == "__main__":
|
243 |
-
demo.queue(
|
|
|
11 |
from openai import OpenAI
|
12 |
import subprocess
|
13 |
from starlette.requests import ClientDisconnect
|
|
|
|
|
14 |
|
15 |
LLAMA_3B_API_ENDPOINT = os.environ.get("LLAMA_3B_API_ENDPOINT")
|
16 |
LLAMA_3B_API_KEY = os.environ.get("LLAMA_3B_API_KEY")
|
|
|
18 |
|
19 |
default_lang = "en"
|
20 |
engines = { default_lang: Model(default_lang) }
|
|
|
|
|
21 |
|
22 |
LANGUAGE_CODES = {
|
23 |
"English": "eng",
|
|
|
120 |
|
121 |
return output
|
122 |
|
123 |
+
def translate_speech(audio_file, target_language):
|
124 |
+
if audio_file is None:
|
125 |
+
return None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
|
127 |
+
language_code = LANGUAGE_CODES[target_language]
|
128 |
+
output_file = "translated_audio.wav"
|
|
|
|
|
|
|
129 |
|
130 |
+
command = [
|
131 |
+
"expressivity_predict",
|
132 |
+
audio_file,
|
133 |
+
"--tgt_lang", language_code,
|
134 |
+
"--model_name", "seamless_expressivity",
|
135 |
+
"--vocoder_name", "vocoder_pretssel",
|
136 |
+
"--gated-model-dir", "models",
|
137 |
+
"--output_path", output_file
|
138 |
+
]
|
139 |
|
140 |
+
subprocess.run(command, check=True)
|
|
|
141 |
|
142 |
+
if os.path.exists(output_file):
|
143 |
+
print(f"File created successfully: {output_file}")
|
144 |
+
return output_file
|
145 |
+
else:
|
146 |
+
print(f"File not found: {output_file}")
|
|
|
|
|
147 |
return None
|
148 |
|
149 |
async def respond(audio, model, seed, target_language):
|
150 |
try:
|
151 |
if audio is None:
|
152 |
+
return None, None, "No input detected."
|
153 |
+
|
154 |
user_input = transcribe(audio)
|
155 |
if not user_input:
|
156 |
+
return None, None, "Could not transcribe audio."
|
157 |
+
|
158 |
if user_input.lower().startswith("please translate"):
|
159 |
+
# Extract the actual content to translate
|
160 |
+
content_to_translate = user_input[len("please translate"):].strip()
|
161 |
+
translated_audio = translate_speech(audio, target_language)
|
162 |
+
return None, translated_audio, f"Translated to {target_language}"
|
163 |
else:
|
164 |
reply = models(user_input, model, seed)
|
165 |
communicate = edge_tts.Communicate(reply, voice="en-US-ChristopherNeural")
|
166 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
|
167 |
tmp_path = tmp_file.name
|
168 |
await communicate.save(tmp_path)
|
169 |
+
return tmp_path, None, "Voice assistant response"
|
170 |
except ClientDisconnect:
|
171 |
print("Client disconnected")
|
172 |
+
return None, None, "Client disconnected. Please try again."
|
173 |
except Exception as e:
|
174 |
print(f"An error occurred: {str(e)}")
|
175 |
+
return None, None, f"An error occurred: {str(e)}"
|
176 |
|
177 |
def clear_history():
|
178 |
global conversation_history
|
179 |
conversation_history = []
|
180 |
+
return None, None, "Conversation history cleared."
|
181 |
|
182 |
with gr.Blocks(css="style.css") as demo:
|
183 |
gr.Markdown("# <center><b>Optimus Prime: Voice Assistant with Translation</b></center>")
|
|
|
185 |
|
186 |
with gr.Row():
|
187 |
with gr.Column(scale=1):
|
|
|
188 |
select = gr.Dropdown([
|
189 |
'Llama 3 8B Service',
|
190 |
'Mixtral 8x7B',
|
|
|
195 |
value="Llama 3 8B Service",
|
196 |
label="Model"
|
197 |
)
|
|
|
|
|
|
|
|
|
|
|
198 |
seed = gr.Slider(
|
199 |
label="Seed",
|
200 |
minimum=0,
|
|
|
203 |
value=0,
|
204 |
visible=False
|
205 |
)
|
206 |
+
target_lang = gr.Dropdown(
|
207 |
+
choices=list(LANGUAGE_CODES.keys()),
|
208 |
+
value="German",
|
209 |
+
label="Target Language for Translation"
|
210 |
+
)
|
211 |
+
input_audio = gr.Audio(label="User Input", sources=["microphone"], type="filepath")
|
212 |
clear_button = gr.Button("Clear Conversation History")
|
213 |
|
214 |
with gr.Column(scale=1):
|
215 |
output_audio = gr.Audio(label="AI Response", type="filepath", interactive=False, autoplay=True)
|
216 |
translated_audio = gr.Audio(label="Translated Audio", type="filepath", interactive=False, autoplay=True)
|
217 |
+
status_message = gr.Textbox(label="Status", interactive=False)
|
218 |
|
219 |
input_audio.change(
|
220 |
fn=respond,
|
221 |
inputs=[input_audio, select, seed, target_lang],
|
222 |
+
outputs=[output_audio, translated_audio, status_message],
|
223 |
)
|
224 |
|
225 |
+
clear_button.click(fn=clear_history, inputs=[], outputs=[output_audio, translated_audio, status_message])
|
226 |
|
227 |
if __name__ == "__main__":
|
228 |
+
demo.queue(max_size=200).launch()
|