Update app.py
Browse files
app.py
CHANGED
@@ -1,34 +1,61 @@
|
|
1 |
import os
|
2 |
import httpx
|
3 |
-
|
|
|
|
|
|
|
4 |
from openai import OpenAI
|
5 |
|
6 |
-
# Initialize
|
7 |
sambanova_client = OpenAI(
|
8 |
-
api_key=os.getenv("key"),
|
|
|
9 |
)
|
10 |
-
|
11 |
-
# Load STT and TTS models
|
12 |
stt_model = get_stt_model()
|
13 |
tts_model = get_tts_model()
|
14 |
|
15 |
-
|
16 |
-
class EchoHandler(StreamHandlerBase):
|
17 |
def __init__(self):
|
18 |
-
super().__init__()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
-
|
21 |
-
prompt = stt_model.stt(audio)
|
22 |
response = sambanova_client.chat.completions.create(
|
23 |
model="mistralai/Mistral-Small-24B-Instruct-2501",
|
24 |
-
messages=[{"role": "user", "content":
|
25 |
max_tokens=200,
|
26 |
)
|
27 |
reply = response.choices[0].message.content
|
28 |
-
for audio_chunk in tts_model.stream_tts_sync(reply):
|
29 |
-
yield audio_chunk
|
30 |
|
31 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
def get_cloudflare_turn_credentials(
|
33 |
turn_key_id=None,
|
34 |
turn_key_api_token=None,
|
@@ -36,20 +63,15 @@ def get_cloudflare_turn_credentials(
|
|
36 |
ttl=600,
|
37 |
client: httpx.AsyncClient | None = None,
|
38 |
):
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
}
|
44 |
-
]
|
45 |
-
}
|
46 |
-
|
47 |
-
# Launch stream with correct handler
|
48 |
stream = Stream(
|
49 |
handler=EchoHandler(),
|
50 |
-
rtc_configuration=get_cloudflare_turn_credentials,
|
51 |
modality="audio",
|
52 |
-
mode="send-receive"
|
|
|
53 |
)
|
54 |
|
55 |
stream.fastphone()
|
|
|
1 |
import os
|
2 |
import httpx
|
3 |
+
import numpy as np
|
4 |
+
from queue import Queue, Empty
|
5 |
+
|
6 |
+
from fastrtc import Stream, StreamHandler, get_stt_model, get_tts_model
|
7 |
from openai import OpenAI
|
8 |
|
9 |
+
# Initialize OpenAI client and on-device models
|
10 |
sambanova_client = OpenAI(
|
11 |
+
api_key=os.getenv("key"),
|
12 |
+
base_url="https://api.deepinfra.com/v1"
|
13 |
)
|
|
|
|
|
14 |
stt_model = get_stt_model()
|
15 |
tts_model = get_tts_model()
|
16 |
|
17 |
+
class EchoHandler(StreamHandler):
|
|
|
18 |
def __init__(self):
|
19 |
+
super().__init__() # uses default sample rates/layouts
|
20 |
+
self.queue: Queue[tuple[int, np.ndarray]] = Queue()
|
21 |
+
|
22 |
+
def start_up(self) -> None:
|
23 |
+
# Optional: warm up models or state here
|
24 |
+
pass
|
25 |
+
|
26 |
+
def receive(self, frame: tuple[int, np.ndarray]) -> None:
|
27 |
+
# frame is (sample_rate, numpy array)
|
28 |
+
sample_rate, audio_array = frame
|
29 |
+
|
30 |
+
# 1) Transcribe speech → text
|
31 |
+
text = stt_model.stt(frame)
|
32 |
|
33 |
+
# 2) Chat completion
|
|
|
34 |
response = sambanova_client.chat.completions.create(
|
35 |
model="mistralai/Mistral-Small-24B-Instruct-2501",
|
36 |
+
messages=[{"role": "user", "content": text}],
|
37 |
max_tokens=200,
|
38 |
)
|
39 |
reply = response.choices[0].message.content
|
|
|
|
|
40 |
|
41 |
+
# 3) Generate TTS chunks and enqueue them
|
42 |
+
for tts_chunk in tts_model.stream_tts_sync(reply):
|
43 |
+
# each tts_chunk is a numpy array of shape (1, N)
|
44 |
+
self.queue.put((sample_rate, tts_chunk))
|
45 |
+
|
46 |
+
def emit(self):
|
47 |
+
try:
|
48 |
+
return self.queue.get_nowait()
|
49 |
+
except Empty:
|
50 |
+
return None # no audio to send right now
|
51 |
+
|
52 |
+
def copy(self) -> "EchoHandler":
|
53 |
+
return EchoHandler()
|
54 |
+
|
55 |
+
def shutdown(self) -> None:
|
56 |
+
# Optional cleanup
|
57 |
+
pass
|
58 |
+
|
59 |
def get_cloudflare_turn_credentials(
|
60 |
turn_key_id=None,
|
61 |
turn_key_api_token=None,
|
|
|
63 |
ttl=600,
|
64 |
client: httpx.AsyncClient | None = None,
|
65 |
):
|
66 |
+
# Replace with your real TURN creds logic
|
67 |
+
return {"iceServers": [{"urls": ["stun:stun.l.google.com:19302"]}]}
|
68 |
+
|
69 |
+
# Wire up the stream with the new handler
|
|
|
|
|
|
|
|
|
|
|
70 |
stream = Stream(
|
71 |
handler=EchoHandler(),
|
|
|
72 |
modality="audio",
|
73 |
+
mode="send-receive",
|
74 |
+
rtc_configuration=get_cloudflare_turn_credentials
|
75 |
)
|
76 |
|
77 |
stream.fastphone()
|