GLorr's picture
Upload folder using huggingface_hub
6c09f76 verified
raw
history blame contribute delete
16.7 kB
# -*- coding: utf-8 -*-
# Copyright 2025 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
## Setup
The gradio-webrtc install fails unless you have ffmpeg@6, on mac:
```
brew uninstall ffmpeg
brew install ffmpeg@6
brew link ffmpeg@6
```
Create a virtual python environment, then install the dependencies for this script:
```
pip install websockets numpy gradio-webrtc "gradio>=5.9.1"
```
If installation fails it may be
Before running this script, ensure the `GOOGLE_API_KEY` environment
```
$ export GOOGLE_API_KEY ='add your key here'
```
You can get an api-key from Google AI Studio (https://aistudio.google.com/apikey)
## Run
To run the script:
```
python gemini_gradio_audio.py
```
On the gradio page (http://127.0.0.1:7860/) click record, and talk, gemini will reply. But note that interruptions
don't work.
"""
import base64
import json
import os
import wave
import itertools
import gradio as gr
import numpy as np
import websockets.sync.client
from gradio_webrtc import StreamHandler, WebRTC
from jinja2 import Template
import threading
import queue
from tools import FUNCTION_MAP, TOOLS
from google.cloud import texttospeech
# logging.basicConfig(
# level=logging.INFO,
# format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
# )
# logger = logging.getLogger(__name__)
with open("questions.json", "r") as f:
questions_dict = json.load(f)
with open("src/prompts/default_prompt.jinja2") as f:
template_str = f.read()
template = Template(template_str)
system_prompt = template.render(questions=json.dumps(questions_dict, indent=4))
print(system_prompt)
# TOOLS = types.GenerateContentConfig(tools=[validate_answer])
__version__ = "0.0.3"
KEY_NAME = "GOOGLE_API_KEY"
# Configuration and Utilities
class GeminiConfig:
"""Configuration settings for Gemini API."""
def __init__(self):
self.api_key = os.getenv(KEY_NAME)
self.host = "generativelanguage.googleapis.com"
self.model = "models/gemini-2.0-flash-exp"
self.ws_url = f"wss://{self.host}/ws/google.ai.generativelanguage.v1alpha.GenerativeService.BidiGenerateContent?key={self.api_key}"
class TTSStreamer:
def __init__(self):
self.client = texttospeech.TextToSpeechClient()
self.text_queue = queue.Queue()
self.audio_queue = queue.Queue()
def start_stream(self):
streaming_config = texttospeech.StreamingSynthesizeConfig(
voice=texttospeech.VoiceSelectionParams(
name="en-US-Journey-D",
language_code="en-US"
)
)
config_request = texttospeech.StreamingSynthesizeRequest(
streaming_config=streaming_config
)
def request_generator():
while True:
try:
text = self.text_queue.get()
if text is None: # Poison pill to stop
break
yield texttospeech.StreamingSynthesizeRequest(
input=texttospeech.StreamingSynthesisInput(text=text)
)
except queue.Empty:
continue
def audio_processor():
responses = self.client.streaming_synthesize(
itertools.chain([config_request], request_generator())
)
print(f"Responses: {responses}")
for response in responses:
self.audio_queue.put(response.audio_content)
self.processor_thread = threading.Thread(target=audio_processor)
self.processor_thread.start()
def send_text(self, text: str):
"""Send text to be synthesized."""
self.text_queue.put(text)
def get_audio(self):
"""Get the next chunk of audio bytes."""
try:
return self.audio_queue.get_nowait()
except queue.Empty:
return None
def stop(self):
"""Stop the streaming synthesis."""
self.text_queue.put(None) # Send poison pill
if self.processor_thread:
self.processor_thread.join()
class AudioProcessor:
"""Handles encoding and decoding of audio data."""
@staticmethod
def encode_audio(data, sample_rate):
"""Encodes audio data to base64."""
encoded = base64.b64encode(data.tobytes()).decode("UTF-8")
return {
"realtimeInput": {
"mediaChunks": [
{
"mimeType": f"audio/pcm;rate={sample_rate}",
"data": encoded,
}
],
},
}
@staticmethod
def process_audio_response(data):
"""Decodes audio data from base64."""
audio_data = base64.b64decode(data)
return np.frombuffer(audio_data, dtype=np.int16)
# Gemini Interaction Handler
class GeminiHandler(StreamHandler):
"""Handles streaming interactions with the Gemini API."""
def __init__(
self,
audio_file=None,
expected_layout="mono",
output_sample_rate=24000,
output_frame_size=480,
) -> None:
super().__init__(
expected_layout,
output_sample_rate,
output_frame_size,
input_sample_rate=24000,
)
self.config = GeminiConfig()
self.ws = None
self.all_output_data = None
self.audio_processor = AudioProcessor()
self.audio_file = audio_file
self.text_buffer = ""
self.tts_engine = None
def copy(self):
"""Creates a copy of the GeminiHandler instance."""
return GeminiHandler(
expected_layout=self.expected_layout,
output_sample_rate=self.output_sample_rate,
output_frame_size=self.output_frame_size,
)
def _initialize_websocket(self):
"""Initializes the WebSocket connection to the Gemini API."""
try:
self.ws = websockets.sync.client.connect(self.config.ws_url, timeout=3000)
setup_request = {
"setup": {
"model": self.config.model,
"tools": [{"functionDeclarations": TOOLS}],
"generationConfig": {"responseModalities": "TEXT"},
"systemInstruction": {
"parts": [{"text": system_prompt}],
"role": "user",
},
}
}
self.ws.send(json.dumps(setup_request))
setup_response = json.loads(self.ws.recv())
print(f"Setup response: {setup_response}")
if self.audio_file:
self.input_audio_file(self.audio_file)
print("Audio file sent")
except websockets.exceptions.WebSocketException as e:
print(f"WebSocket connection failed: {str(e)}")
self.ws = None
except Exception as e:
print(f"Setup failed: {str(e)}")
self.ws = None
def input_audio_file(self, audio_file):
"""Processes an audio file and sends it to the Gemini API."""
try:
with wave.open(audio_file, "rb") as wf:
data = wf.readframes(wf.getnframes())
self.receive((wf.getframerate(), np.frombuffer(data, dtype=np.int16)))
except Exception as e:
print(f"Error in input_audio_file: {str(e)}")
def receive(self, frame: tuple[int, np.ndarray]) -> None:
"""Receives audio/video data, encodes it, and sends it to the Gemini API."""
try:
if not self.ws:
self._initialize_websocket()
sample_rate, array = frame
message = {"realtimeInput": {"mediaChunks": []}}
if sample_rate > 0 and array is not None:
array = array.squeeze()
audio_data = self.audio_processor.encode_audio(
array, self.output_sample_rate
)
message["realtimeInput"]["mediaChunks"].append(
{
"mimeType": f"audio/pcm;rate={self.output_sample_rate}",
"data": audio_data["realtimeInput"]["mediaChunks"][0]["data"],
}
)
if message["realtimeInput"]["mediaChunks"]:
self.ws.send(json.dumps(message))
except Exception as e:
print(f"Error in receive: {str(e)}")
if self.ws:
self.ws.close()
self.ws = None
def handle_tool_call(self, tool_call):
print(" ", tool_call)
for fc in tool_call["functionCalls"]:
print(f"Function call: {fc}")
# Call the function
try:
result = {"output": FUNCTION_MAP[fc["name"]](**fc["args"])}
except Exception as e:
result = {"error": str(e)}
# Send the response back
msg = {
"tool_response": {
"function_responses": [
{"id": fc["id"], "name": fc["name"], "response": result}
]
}
}
print(f"function response: {msg}")
self.ws.send(json.dumps(msg))
def _output_data(self, audio_array):
"""Processes audio output data from the WebSocket response."""
if self.all_output_data is None:
self.all_output_data = audio_array
else:
self.all_output_data = np.concatenate((self.all_output_data, audio_array))
while self.all_output_data.shape[-1] >= self.output_frame_size:
yield (
self.output_sample_rate,
self.all_output_data[: self.output_frame_size].reshape(1, -1),
)
self.all_output_data = self.all_output_data[self.output_frame_size :]
def _process_server_content(self, content):
"""Processes audio output data from the WebSocket response."""
if respone := content.get("modelTurn", {}):
if parts:= respone.get("parts"):
for part in parts:
print(f"Part: {part}")
data = part.get("inlineData", {}).get("data", "")
if data:
audio_array = self.audio_processor.process_audio_response(data)
yield from self._output_data(audio_array)
text = part.get("text", "")
if text:
self.text_buffer += text
# audio_array = self._text_to_audio(text)
# yield from self._output_data(audio_array)
# # self.text_buffer += text
# Check if the turn is complete and process the text buffer into audio
if content.get("turnComplete"):
if self.text_buffer:
audio_array = self._text_to_audio(self.text_buffer)
yield from self._output_data(audio_array)
self.text_buffer = ""
def _text_to_audio(self, text: str) -> np.ndarray:
"""Convert text to audio using Google Cloud TTS streaming."""
client = texttospeech.TextToSpeechClient()
# Configure synthesis
synthesis_input = texttospeech.SynthesisInput(text=text)
voice = texttospeech.VoiceSelectionParams(
name="en-IN-Chirp-HD-O",
language_code="en-IN"
)
audio_config = texttospeech.AudioConfig(
audio_encoding=texttospeech.AudioEncoding.LINEAR16
)
# Get response in a single request
try:
response = client.synthesize_speech(
input=synthesis_input,
voice=voice,
audio_config=audio_config
)
return np.frombuffer(response.audio_content, dtype=np.int16)
except Exception as e:
print(f"Error in speech synthesis: {e}")
return np.array([], dtype=np.int16)
def generator(self):
"""Generates audio output from the WebSocket stream."""
while True:
if not self.ws:
print("WebSocket not connected")
yield None
continue
try:
message = self.ws.recv(timeout=30)
msg = json.loads(message)
# {'serverContent': {'modelTurn': {'parts': [{'text': 'Hello'}]}}}
# {'serverContent': {'modelTurn': {'parts': [{'text': ', good morning! Thank you for taking my call. My name is [Your'}]}}}
# {'serverContent': {'modelTurn': {'parts': [{'text': " Name] and I'm a technical recruiter. I'm conducting a quick"}]}}}
# {'serverContent': {'modelTurn': {'parts': [{'text': ' initial screening, is that okay with you?\n'}]}}}
# {'serverContent': {'turnComplete': True}}
if "serverContent" in msg:
content = msg["serverContent"]
yield from self._process_server_content(content)
elif "toolCall" in msg:
yield from self.handle_tool_call(msg["toolCall"])
except TimeoutError:
print("Timeout waiting for server response")
yield None
except Exception:
yield None
def emit(self) -> tuple[int, np.ndarray] | None:
"""Emits the next audio chunk from the generator."""
if not self.ws:
return None
if not hasattr(self, "_generator"):
self._generator = self.generator()
try:
return next(self._generator)
except StopIteration:
self.reset()
return None
def reset(self) -> None:
"""Resets the generator and output data."""
if hasattr(self, "_generator"):
delattr(self, "_generator")
self.all_output_data = None
def shutdown(self) -> None:
"""Closes the WebSocket connection."""
if self.ws:
self.ws.close()
def check_connection(self):
"""Checks if the WebSocket connection is active."""
try:
if not self.ws or self.ws.closed:
self._initialize_websocket()
return True
except Exception as e:
print(f"Connection check failed: {str(e)}")
return False
def update_answers():
with open("answers.json", "r") as f:
return json.load(f)
# Main Gradio Interface
def registry(name: str, token: str | None = None, **kwargs):
"""Sets up and returns the Gradio interface."""
api_key = token or os.environ.get(KEY_NAME)
if not api_key:
raise ValueError(f"{KEY_NAME} environment variable is not set.")
interface = gr.Blocks()
with interface:
with gr.Tabs():
with gr.TabItem("Voice Chat"):
gr.HTML(
"""
<div style='text-align: left'>
<h1>ML6 Voice Demo - Function Calling and Custom Output Voice</h1>
</div>
"""
)
gemini_handler = GeminiHandler()
# gemini_handler = ThreeStepHandler()
with gr.Row():
audio = WebRTC(
label="Voice Chat", modality="audio", mode="send-receive"
)
# Add display components for questions and answers
with gr.Row():
with gr.Column():
gr.JSON(
label="Questions",
value=questions_dict,
)
with gr.Column():
gr.JSON(update_answers, label="Collected Answers", every=1)
audio.stream(
gemini_handler,
inputs=[audio], # Add audio_file to inputs
outputs=[audio],
time_limit=600,
concurrency_limit=10,
)
return interface
# Launch the Gradio interface
gr.load(
name="gemini-2.0-flash-exp",
src=registry,
).launch()