File size: 7,142 Bytes
dff5fe4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
import os
import time
import gradio as gr
import numpy as np
from dotenv import load_dotenv
from elevenlabs import ElevenLabs
from fastrtc import (
    Stream,
    get_stt_model,
    ReplyOnPause,
    AdditionalOutputs
)

import requests
import io
import soundfile as sf
from gtts import gTTS
import re
import inspect

from deepseek import DeepSeekAPI

# Load environment variables
load_dotenv()

# Initialize clients
elevenlabs_client = ElevenLabs(api_key=os.getenv("ELEVENLABS_API_KEY"))
stt_model = get_stt_model()
deepseek_client = DeepSeekAPI(api_key=os.getenv("DEEPSEEK_API_KEY"))

# Add this debug code temporarily to see what methods are available:
print(dir(deepseek_client))

def response(

    audio: tuple[int, np.ndarray],

    chatbot: list[dict] | None = None,

):
    chatbot = chatbot or []
    messages = [{"role": d["role"], "content": d["content"]} for d in chatbot]
    
    # Convert speech to text
    text = stt_model.stt(audio)
    print("prompt:", text)
    
    # Add user message to chat
    chatbot.append({"role": "user", "content": text})
    yield AdditionalOutputs(chatbot)
    
    # Get AI response
    messages.append({"role": "user", "content": text})
    response_text = get_deepseek_response(messages)
    
    # Add AI response to chat
    chatbot.append({"role": "assistant", "content": response_text})
    
    # Convert response to speech
    for audio_data in text_to_speech(response_text):
        if audio_data:
            yield audio_data
    
    yield AdditionalOutputs(chatbot)

# Create Gradio interface
chatbot = gr.Chatbot(type="messages")
stream = Stream(
    modality="audio",
    mode="send-receive",
    handler=ReplyOnPause(response, input_sample_rate=16000),
    additional_outputs_handler=lambda a, b: b,
    additional_inputs=[chatbot],
    additional_outputs=[chatbot],
    ui_args={"title": "LLM Voice Chat (Powered by DeepSeek & ElevenLabs)"}
)

# Create FastAPI app and mount stream
from fastapi import FastAPI
app = FastAPI()
app = gr.mount_gradio_app(app, stream.ui, path="/")
stream.mount(app)  # Mount the stream for telephone/fastphone integration

# Update the chat completion part based on available methods:
# We'll use direct HTTP requests as a fallback since the API structure is unclear:
def get_deepseek_response(messages):
    url = "https://api.deepseek.com/v1/chat/completions"
    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {os.getenv('DEEPSEEK_API_KEY')}"
    }
    payload = {
        "model": "deepseek-chat",
        "messages": messages,
        "temperature": 0.7,
        "max_tokens": 512
    }
    response = requests.post(url, json=payload, headers=headers)
    
    # Check for error response
    if response.status_code != 200:
        print(f"DeepSeek API error: {response.status_code} - {response.text}")
        return "I'm sorry, I encountered an error processing your request."
        
    response_json = response.json()
    return response_json["choices"][0]["message"]["content"]

# Make sure that the text_to_speech function is completely replaced and gTTS is explicitly using US English
def text_to_speech(text):
    """Convert text to speech using Google TTS with sentence-by-sentence processing"""
    try:
        # Split text into sentences for faster perceived response
        sentences = re.split(r'(?<=[.!?])\s+', text)
        
        for sentence in sentences:
            if not sentence.strip():
                continue
                
            # Process each sentence separately
            mp3_fp = io.BytesIO()
            
            # Force US English - be explicit about it
            print(f"Using gTTS with en-us locale for sentence: {sentence[:20]}...")
            tts = gTTS(text=sentence, lang='en-us', tld='com', slow=False)
            tts.write_to_fp(mp3_fp)
            mp3_fp.seek(0)
            
            # Process audio data
            data, samplerate = sf.read(mp3_fp)
            
            # Convert to mono if stereo
            if len(data.shape) > 1 and data.shape[1] > 1:
                data = data[:, 0]
            
            # Resample to 24000 Hz if needed
            if samplerate != 24000:
                data = np.interp(
                    np.linspace(0, len(data), int(len(data) * 24000 / samplerate)),
                    np.arange(len(data)),
                    data
                )
            
            # Convert to 16-bit integers
            data = (data * 32767).astype(np.int16)
            
            # Ensure buffer size is even
            if len(data) % 2 != 0:
                data = np.append(data, [0])
            
            # Reshape and yield in chunks
            chunk_size = 4800
            for i in range(0, len(data), chunk_size):
                chunk = data[i:i+chunk_size]
                if len(chunk) > 0:
                    if len(chunk) % 2 != 0:
                        chunk = np.append(chunk, [0])
                    chunk = chunk.reshape(1, -1)
                    yield (24000, chunk)
    except Exception as e:
        print(f"Exception in text_to_speech: {e}")
        yield None

# Add this debug statement AFTER the function definition
print("text_to_speech function:", inspect.getsource(text_to_speech))

if __name__ == "__main__":
    os.environ["GRADIO_SSR_MODE"] = "false"
    
    # Check FastRTC version
    import fastrtc
    print(f"FastRTC version: {fastrtc.__version__ if hasattr(fastrtc, '__version__') else 'unknown'}")
    
    # Try running fastphone with additional diagnostic
    print("Starting phone service - attempting to inspect fastphone method...")
    import inspect
    print(f"FastPhone signature: {inspect.signature(stream.fastphone) if hasattr(stream, 'fastphone') else 'Not available'}")
    
    try:
        # Fix: Use keyword argument instead of positional
        phone_service = stream.fastphone(
            token=os.getenv("HF_TOKEN"),
            host="127.0.0.1",
            port=8000,
            share_server_tls_certificate=True  # Use keyword argument format
        )
        print("Phone service started successfully")
    except Exception as e:
        print(f"Error starting phone service: {e}")
        print("Falling back to web interface...")
        # Launch with web interface as fallback
        stream.ui.launch(server_port=7860)

# Remove or comment out the following lines:
# !pip install -q torch==2.0.1 torchaudio==2.0.2 gradio requests soundfile huggingface_hub
# !wget -q https://github.com/seasalt-ai/csm/archive/refs/heads/main.zip
# !unzip -q main.zip
# !mv csm-main csm
# !cd csm && pip install -e .
# 
# # Set up directories
# import os
# import sys
# sys.path.append("/content/csm")
# voice_samples_dir = "/content/csm_voice_samples"
# output_dir = "/content/csm_output"
# os.makedirs(voice_samples_dir, exist_ok=True)
# os.makedirs(output_dir, exist_ok=True)
# 
# print("✅ Dependencies installed!")