Spaces:
Running
Running
Upload folder using huggingface_hub
Browse files
app.py
CHANGED
|
@@ -1,66 +1,21 @@
|
|
| 1 |
-
import os
|
| 2 |
-
import time
|
| 3 |
-
|
| 4 |
import numpy as np
|
| 5 |
-
from dotenv import load_dotenv
|
| 6 |
from fastapi import FastAPI
|
| 7 |
from fastapi.responses import RedirectResponse
|
| 8 |
-
from fastrtc import
|
| 9 |
-
ReplyOnPause,
|
| 10 |
-
Stream,
|
| 11 |
-
get_stt_model,
|
| 12 |
-
get_tts_model,
|
| 13 |
-
)
|
| 14 |
from gradio.utils import get_space
|
| 15 |
-
from numpy.typing import NDArray
|
| 16 |
-
from openai import OpenAI
|
| 17 |
-
|
| 18 |
-
load_dotenv()
|
| 19 |
-
|
| 20 |
-
sambanova_client = OpenAI(
|
| 21 |
-
api_key=os.getenv("SAMBANOVA_API_KEY"), base_url="https://api.sambanova.ai/v1"
|
| 22 |
-
)
|
| 23 |
-
|
| 24 |
-
stt_model = get_stt_model()
|
| 25 |
-
tts_model = get_tts_model()
|
| 26 |
|
| 27 |
-
chat_history = [
|
| 28 |
-
{
|
| 29 |
-
"role": "system",
|
| 30 |
-
"content": (
|
| 31 |
-
"You are a helpful assistant having a spoken conversation."
|
| 32 |
-
"Please keep your answers short and concise."
|
| 33 |
-
),
|
| 34 |
-
}
|
| 35 |
-
]
|
| 36 |
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
chat_history.append({"role": "user", "content": prompt})
|
| 42 |
-
start_time = time.time()
|
| 43 |
-
response = sambanova_client.chat.completions.create(
|
| 44 |
-
model="Meta-Llama-3.2-3B-Instruct",
|
| 45 |
-
messages=chat_history,
|
| 46 |
-
max_tokens=200,
|
| 47 |
-
)
|
| 48 |
-
end_time = time.time()
|
| 49 |
-
print("time taken inference", end_time - start_time)
|
| 50 |
-
prompt = response.choices[0].message.content
|
| 51 |
-
chat_history.append({"role": "assistant", "content": prompt})
|
| 52 |
-
start_time = time.time()
|
| 53 |
-
for audio_chunk in tts_model.stream_tts_sync(prompt):
|
| 54 |
-
yield audio_chunk
|
| 55 |
-
end_time = time.time()
|
| 56 |
-
print("time taken tts", end_time - start_time)
|
| 57 |
|
| 58 |
|
| 59 |
stream = Stream(
|
| 60 |
-
handler=ReplyOnPause(
|
| 61 |
modality="audio",
|
| 62 |
mode="send-receive",
|
| 63 |
-
rtc_configuration=
|
| 64 |
concurrency_limit=20 if get_space() else None,
|
| 65 |
)
|
| 66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import numpy as np
|
|
|
|
| 2 |
from fastapi import FastAPI
|
| 3 |
from fastapi.responses import RedirectResponse
|
| 4 |
+
from fastrtc import ReplyOnPause, Stream, get_twilio_turn_credentials
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
from gradio.utils import get_space
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
|
| 8 |
+
def detection(audio: tuple[int, np.ndarray]):
|
| 9 |
+
# Implement any iterator that yields audio
|
| 10 |
+
# See "LLM Voice Chat" for a more complete example
|
| 11 |
+
yield audio
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
|
| 14 |
stream = Stream(
|
| 15 |
+
handler=ReplyOnPause(detection),
|
| 16 |
modality="audio",
|
| 17 |
mode="send-receive",
|
| 18 |
+
rtc_configuration=get_twilio_turn_credentials() if get_space() else None,
|
| 19 |
concurrency_limit=20 if get_space() else None,
|
| 20 |
)
|
| 21 |
|