Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,300 +1,24 @@
|
|
1 |
-
import os
|
2 |
-
import base64
|
3 |
-
import json
|
4 |
-
import numpy as np
|
5 |
import gradio as gr
|
6 |
-
|
7 |
-
|
8 |
-
#PeerConnection,
|
9 |
-
DataChannel,
|
10 |
-
MediaStreamTrack,
|
11 |
-
AudioFrame,
|
12 |
-
VideoFrame,
|
13 |
-
) # Import FastRTC components
|
14 |
-
from aiortc.contrib.media import MediaPlayer, MediaRelay
|
15 |
-
import asyncio
|
16 |
-
|
17 |
-
__version__ = "0.0.3"
|
18 |
-
|
19 |
-
# KEY_NAME = "AIzaSyCWPviRPxj8IMLaijLGbRIsio3dO2rp3rU" # Best practice: Keep API keys out of the main code. Use environment variables.
|
20 |
-
|
21 |
-
|
22 |
-
# Configuration and Utilities
|
23 |
-
class GeminiConfig:
|
24 |
-
"""Configuration settings for Gemini API."""
|
25 |
-
|
26 |
-
def __init__(self):
|
27 |
-
self.api_key = os.environ.get("KEY_NAME") # Use a more descriptive name
|
28 |
-
if not self.api_key:
|
29 |
-
raise ValueError("GEMINI_API_KEY environment variable is not set.")
|
30 |
-
self.host = "generativelanguage.googleapis.com"
|
31 |
-
self.model = "models/gemini-2.0-flash-exp"
|
32 |
-
# FastRTC doesn't use WebSockets directly in the same way. We'll handle the API calls differently.
|
33 |
-
self.base_url = f"https://{self.host}/v1alpha/{self.model}:streamGenerateContent?key={self.api_key}"
|
34 |
-
|
35 |
-
|
36 |
-
class AudioProcessor:
|
37 |
-
"""Handles encoding and decoding of audio data."""
|
38 |
-
|
39 |
-
@staticmethod
|
40 |
-
def encode_audio(data: np.ndarray, sample_rate: int) -> str:
|
41 |
-
"""Encodes audio data to base64."""
|
42 |
-
# Ensure data is in the correct format (int16)
|
43 |
-
if data.dtype != np.int16:
|
44 |
-
data = data.astype(np.int16)
|
45 |
-
encoded = base64.b64encode(data.tobytes()).decode("UTF-8")
|
46 |
-
return encoded
|
47 |
-
|
48 |
-
@staticmethod
|
49 |
-
def process_audio_response(data: str) -> np.ndarray:
|
50 |
-
"""Decodes audio data from base64."""
|
51 |
-
audio_data = base64.b64decode(data)
|
52 |
-
return np.frombuffer(audio_data, dtype=np.int16)
|
53 |
-
|
54 |
-
|
55 |
-
# We don't need a StreamHandler in the same way with FastRTC. We'll handle streaming directly.
|
56 |
-
class GeminiHandler:
|
57 |
-
"""Handles interactions with the Gemini API."""
|
58 |
-
|
59 |
-
def __init__(self, output_sample_rate=24000, output_frame_size=480):
|
60 |
-
self.config = GeminiConfig()
|
61 |
-
self.audio_processor = AudioProcessor()
|
62 |
-
self.output_sample_rate = output_sample_rate
|
63 |
-
self.output_frame_size = output_frame_size
|
64 |
-
self.all_output_data = None
|
65 |
-
self.pc = None # PeerConnection
|
66 |
-
self.dc = None # DataChannel
|
67 |
-
self.audio_track = None
|
68 |
-
self._audio_buffer = []
|
69 |
-
self.relay = MediaRelay()
|
70 |
-
|
71 |
-
async def _send_audio_to_gemini(self, encoded_audio: str):
|
72 |
-
"""Sends audio data to the Gemini API and processes the response."""
|
73 |
-
headers = {"Content-Type": "application/json"}
|
74 |
-
payload = {
|
75 |
-
"contents": [
|
76 |
-
{
|
77 |
-
"parts": [
|
78 |
-
{
|
79 |
-
"text": "Respond to the audio with audio."
|
80 |
-
}, # Initial prompt, can be adjusted
|
81 |
-
{"inline_data": {"mime_type": "audio/pcm;rate=24000", "data": encoded_audio}},
|
82 |
-
]
|
83 |
-
}
|
84 |
-
]
|
85 |
-
}
|
86 |
-
# Use aiohttp for asynchronous HTTP requests
|
87 |
-
import aiohttp
|
88 |
-
|
89 |
-
async with aiohttp.ClientSession() as session:
|
90 |
-
async with session.post(
|
91 |
-
self.config.base_url, headers=headers, data=json.dumps(payload)
|
92 |
-
) as response:
|
93 |
-
if response.status != 200:
|
94 |
-
print(f"Error: Gemini API returned status {response.status}")
|
95 |
-
print(await response.text())
|
96 |
-
return
|
97 |
-
|
98 |
-
async for line in response.content:
|
99 |
-
try:
|
100 |
-
line = line.strip()
|
101 |
-
if not line:
|
102 |
-
continue
|
103 |
-
# Responses are chunked, often with multiple JSON objects per chunk. Handle that.
|
104 |
-
for chunk in line.decode("utf-8").split("\n"):
|
105 |
-
if not chunk.strip():
|
106 |
-
continue
|
107 |
-
try:
|
108 |
-
data = json.loads(chunk)
|
109 |
-
except json.JSONDecodeError:
|
110 |
-
print(f"JSONDecodeError: {chunk}")
|
111 |
-
continue
|
112 |
-
|
113 |
-
if "candidates" in data:
|
114 |
-
for candidate in data["candidates"]:
|
115 |
-
for part in candidate.get("content", {}).get("parts", []):
|
116 |
-
if "inlineData" in part:
|
117 |
-
audio_data = part["inlineData"].get("data", "")
|
118 |
-
if audio_data:
|
119 |
-
await self._process_server_audio(audio_data)
|
120 |
-
|
121 |
-
except Exception as e:
|
122 |
-
print(f"Error processing response chunk: {e}")
|
123 |
-
|
124 |
-
async def _process_server_audio(self, audio_data: str):
|
125 |
-
"""Processes and buffers audio data received from the server."""
|
126 |
-
audio_array = self.audio_processor.process_audio_response(audio_data)
|
127 |
-
if self.all_output_data is None:
|
128 |
-
self.all_output_data = audio_array
|
129 |
-
else:
|
130 |
-
self.all_output_data = np.concatenate((self.all_output_data, audio_array))
|
131 |
-
|
132 |
-
while self.all_output_data.shape[-1] >= self.output_frame_size:
|
133 |
-
frame = AudioFrame(
|
134 |
-
samples=self.output_frame_size,
|
135 |
-
sample_rate=self.output_sample_rate,
|
136 |
-
layout="mono", # mono channel
|
137 |
-
data=self.all_output_data[: self.output_frame_size].tobytes()
|
138 |
-
)
|
139 |
-
self.all_output_data = self.all_output_data[self.output_frame_size:]
|
140 |
-
if self.audio_track:
|
141 |
-
await self.audio_track.emit(frame)
|
142 |
-
|
143 |
-
|
144 |
-
async def on_track(self, track):
|
145 |
-
"""Handles incoming media tracks."""
|
146 |
-
print(f"Track received: {track.kind}")
|
147 |
-
if track.kind == "audio":
|
148 |
-
self.audio_track = track # Store the audio track
|
149 |
-
|
150 |
-
@track.on("frame")
|
151 |
-
async def on_frame(frame):
|
152 |
-
# Process received audio frames
|
153 |
-
if isinstance(frame, AudioFrame):
|
154 |
-
try:
|
155 |
-
# Convert the frame data to a NumPy array
|
156 |
-
audio_data = np.frombuffer(frame.data, dtype=np.int16)
|
157 |
-
# Encode the audio and send it to Gemini
|
158 |
-
encoded_audio = self.audio_processor.encode_audio(
|
159 |
-
audio_data, frame.sample_rate
|
160 |
-
) # Pass sample rate
|
161 |
-
await self._send_audio_to_gemini(encoded_audio)
|
162 |
-
except Exception as e:
|
163 |
-
print(f"Error processing audio frame: {e}")
|
164 |
-
|
165 |
-
async def on_datachannel(self, channel):
|
166 |
-
"""Handles data channel events (not used in this example, but good practice)."""
|
167 |
-
self.dc = channel
|
168 |
-
print("Data channel created")
|
169 |
-
|
170 |
-
@channel.on("message")
|
171 |
-
async def on_message(message):
|
172 |
-
print(f"Received message: {message}")
|
173 |
-
|
174 |
-
async def connect(self):
|
175 |
-
"""Establishes the PeerConnection."""
|
176 |
-
self.pc = PeerConnection()
|
177 |
-
self.pc.on("track", self.on_track)
|
178 |
-
self.pc.on("datachannel", self.on_datachannel)
|
179 |
-
|
180 |
-
# Create a local audio track to send data
|
181 |
-
self.local_audio_player = MediaPlayer("default", format="avfoundation", options={"channels": "1", "sample_rate": str(self.output_sample_rate)})
|
182 |
-
self.local_audio = self.relay.subscribe(self.local_audio_player.audio)
|
183 |
-
self.pc.addTrack(self.local_audio)
|
184 |
-
|
185 |
-
# Add a data channel (optional, but good practice)
|
186 |
-
self.dc = self.pc.createDataChannel("data")
|
187 |
-
|
188 |
-
# Create an offer and set local description
|
189 |
-
offer = await self.pc.createOffer()
|
190 |
-
await self.pc.setLocalDescription(offer)
|
191 |
-
print("PeerConnection established")
|
192 |
-
return self.pc.localDescription
|
193 |
-
|
194 |
-
async def set_remote_description(self, sdp, type):
|
195 |
-
"""Sets the remote description."""
|
196 |
-
from aiortc import RTCSessionDescription
|
197 |
-
|
198 |
-
await self.pc.setRemoteDescription(RTCSessionDescription(sdp=sdp, type=type))
|
199 |
-
print("Remote description set")
|
200 |
-
|
201 |
-
if self.pc.remoteDescription.type == "offer":
|
202 |
-
answer = await self.pc.createAnswer()
|
203 |
-
await self.pc.setLocalDescription(answer)
|
204 |
-
return self.pc.localDescription
|
205 |
-
|
206 |
-
async def add_ice_candidate(self, candidate, sdpMid, sdpMLineIndex):
|
207 |
-
"""Adds an ICE candidate."""
|
208 |
-
from aiortc import RTCIceCandidate
|
209 |
-
|
210 |
-
if candidate:
|
211 |
-
try:
|
212 |
-
ice_candidate = RTCIceCandidate(
|
213 |
-
candidate=candidate, sdpMid=sdpMid, sdpMLineIndex=sdpMLineIndex
|
214 |
-
)
|
215 |
-
await self.pc.addIceCandidate(ice_candidate)
|
216 |
-
print("ICE candidate added")
|
217 |
-
except Exception as e:
|
218 |
-
print(f"Error adding ICE candidate: {e}")
|
219 |
-
|
220 |
-
def shutdown(self):
|
221 |
-
"""Closes the PeerConnection."""
|
222 |
-
if self.pc:
|
223 |
-
asyncio.create_task(self.pc.close()) # Close in the background
|
224 |
-
self.pc = None
|
225 |
-
print("PeerConnection closed")
|
226 |
-
|
227 |
-
|
228 |
-
# Gradio Interface
|
229 |
-
async def registry(
|
230 |
-
name: str,
|
231 |
-
token: str | None = None,
|
232 |
-
**kwargs,
|
233 |
-
):
|
234 |
-
"""Sets up and returns the Gradio interface."""
|
235 |
-
gemini_handler = GeminiHandler()
|
236 |
-
|
237 |
-
async def connect_webrtc(sdp, type, candidates):
|
238 |
-
"""Connects to the WebRTC client and handles ICE candidates."""
|
239 |
-
if gemini_handler.pc is None:
|
240 |
-
local_description = await gemini_handler.connect()
|
241 |
-
if local_description:
|
242 |
-
yield json.dumps(
|
243 |
-
{
|
244 |
-
"sdp": local_description.sdp,
|
245 |
-
"type": local_description.type,
|
246 |
-
"candidates": [],
|
247 |
-
}
|
248 |
-
) # Return initial SDP
|
249 |
-
if sdp and type:
|
250 |
-
answer = await gemini_handler.set_remote_description(sdp, type)
|
251 |
-
if answer:
|
252 |
-
yield json.dumps({"sdp": answer.sdp, "type": answer.type, "candidates": []})
|
253 |
-
|
254 |
-
for candidate in candidates:
|
255 |
-
if candidate and candidate.get("candidate"):
|
256 |
-
await gemini_handler.add_ice_candidate(
|
257 |
-
candidate["candidate"], candidate.get("sdpMid"), candidate.get("sdpMLineIndex")
|
258 |
-
)
|
259 |
-
yield json.dumps({"sdp": "", "type": "", "candidates": []}) # Signal completion
|
260 |
-
|
261 |
-
interface = gr.Blocks()
|
262 |
-
with interface:
|
263 |
-
with gr.Tabs():
|
264 |
-
with gr.TabItem("Voice Chat"):
|
265 |
-
gr.HTML(
|
266 |
-
"""
|
267 |
-
<div style='text-align: left'>
|
268 |
-
<h1>Gemini API Voice Chat</h1>
|
269 |
-
</div>
|
270 |
-
"""
|
271 |
-
)
|
272 |
-
with gr.Row():
|
273 |
-
webrtc_out = gr.JSON(label="WebRTC JSON")
|
274 |
-
|
275 |
-
# Use the built-in WebRTC component, but without automatic streaming.
|
276 |
-
webrtc = gr.WebRTC(
|
277 |
-
value={"sdp": "", "type": "", "candidates": []},
|
278 |
-
interactive=True,
|
279 |
-
label="Voice Chat",
|
280 |
-
)
|
281 |
-
|
282 |
-
connect_button = gr.Button("Connect")
|
283 |
-
connect_button.click(
|
284 |
-
connect_webrtc,
|
285 |
-
inputs=[
|
286 |
-
webrtc
|
287 |
-
], # Pass the WebRTC component's value (SDP, type, candidates)
|
288 |
-
outputs=[webrtc_out], # show the webrtc connection data
|
289 |
-
)
|
290 |
-
|
291 |
-
return interface
|
292 |
-
|
293 |
-
# Launch the Gradio interface
|
294 |
-
async def main():
|
295 |
-
interface = await registry(name="gemini-2.0-flash-exp")
|
296 |
-
interface.queue() # Enable queuing for better concurrency
|
297 |
-
await interface.launch()
|
298 |
|
299 |
-
|
300 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
from PyPDF2 import PdfReader
|
3 |
+
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
+
def process_pdf(file):
|
6 |
+
# Read the PDF content
|
7 |
+
pdf_reader = PdfReader(file.name)
|
8 |
+
text = ""
|
9 |
+
for page in pdf_reader.pages:
|
10 |
+
text += page.extract_text()
|
11 |
+
return text
|
12 |
+
|
13 |
+
with gr.Blocks() as demo:
|
14 |
+
gr.Markdown("### File upload", elem_classes="tab-header")
|
15 |
+
with gr.Row():
|
16 |
+
text_output = gr.Textbox(label="text")
|
17 |
+
file_input = gr.File(label="Wähle eine PDF-Datei aus", type="filepath")
|
18 |
+
upload_output = gr.Textbox(label="Upload Status")
|
19 |
+
with gr.Row():
|
20 |
+
submit_button = gr.Button("upload")
|
21 |
+
submit_button.click(process_pdf, inputs=file_input, outputs=text_output
|
22 |
+
|
23 |
+
demo.launch())
|
24 |
+
|