Create sppech_input_interim.py
Browse files- sppech_input_interim.py +194 -0
sppech_input_interim.py
ADDED
@@ -0,0 +1,194 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
Hugging Face's logo
|
2 |
+
Hugging Face
|
3 |
+
Search models, datasets, users...
|
4 |
+
Models
|
5 |
+
Datasets
|
6 |
+
Spaces
|
7 |
+
Posts
|
8 |
+
Docs
|
9 |
+
Enterprise
|
10 |
+
Pricing
|
11 |
+
|
12 |
+
|
13 |
+
|
14 |
+
Spaces:
|
15 |
+
|
16 |
+
DrishtiSharma
|
17 |
+
/
|
18 |
+
chat-w-docs-via-speech-or-text
|
19 |
+
|
20 |
+
|
21 |
+
like
|
22 |
+
1
|
23 |
+
|
24 |
+
Logs
|
25 |
+
App
|
26 |
+
Files
|
27 |
+
Community
|
28 |
+
Settings
|
29 |
+
chat-w-docs-via-speech-or-text
|
30 |
+
/
|
31 |
+
app.py
|
32 |
+
|
33 |
+
DrishtiSharma's picture
|
34 |
+
DrishtiSharma
|
35 |
+
Update app.py
|
36 |
+
230a23e
|
37 |
+
verified
|
38 |
+
3 minutes ago
|
39 |
+
raw
|
40 |
+
|
41 |
+
Copy download link
|
42 |
+
history
|
43 |
+
blame
|
44 |
+
edit
|
45 |
+
delete
|
46 |
+
|
47 |
+
5.38 kB
|
48 |
+
#ref: https://www.youtube.com/watch?v=3ZDVmzlM6Nc
|
49 |
+
|
50 |
+
import os
|
51 |
+
import chromadb
|
52 |
+
import streamlit as st
|
53 |
+
from langchain_huggingface import HuggingFaceEmbeddings
|
54 |
+
from langchain_chroma import Chroma
|
55 |
+
from langchain_groq import ChatGroq
|
56 |
+
from langchain.memory import ConversationBufferMemory
|
57 |
+
from langchain.chains import ConversationalRetrievalChain
|
58 |
+
from PyPDF2 import PdfReader
|
59 |
+
from groq import Groq
|
60 |
+
from streamlit_webrtc import webrtc_streamer, AudioProcessorBase, WebRtcMode
|
61 |
+
import av
|
62 |
+
|
63 |
+
# Clear ChromaDB cache to fix tenant issue
|
64 |
+
chromadb.api.client.SharedSystemClient.clear_system_cache()
|
65 |
+
|
66 |
+
# Ensure required environment variables are set
|
67 |
+
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
|
68 |
+
if not GROQ_API_KEY:
|
69 |
+
st.error("GROQ_API_KEY is not set. Please configure it in environment variables.")
|
70 |
+
st.stop()
|
71 |
+
|
72 |
+
# Initialize Groq Client for transcription and LLM
|
73 |
+
groq_client = Groq(api_key=GROQ_API_KEY)
|
74 |
+
llm = ChatGroq(model="llama-3.1-70b-versatile", temperature=0, groq_api_key=GROQ_API_KEY)
|
75 |
+
|
76 |
+
# Function to process PDFs and set up the vectorstore
|
77 |
+
def process_and_store_pdfs(uploaded_files):
|
78 |
+
texts = []
|
79 |
+
for uploaded_file in uploaded_files:
|
80 |
+
reader = PdfReader(uploaded_file)
|
81 |
+
for page in reader.pages:
|
82 |
+
texts.append(page.extract_text())
|
83 |
+
|
84 |
+
embeddings = HuggingFaceEmbeddings()
|
85 |
+
vectorstore = Chroma.from_texts(texts, embedding=embeddings, persist_directory="vector_db_dir")
|
86 |
+
return vectorstore
|
87 |
+
|
88 |
+
# Function to set up the chat chain
|
89 |
+
def chat_chain(vectorstore):
|
90 |
+
retriever = vectorstore.as_retriever()
|
91 |
+
memory = ConversationBufferMemory(output_key="answer", memory_key="chat_history", return_messages=True)
|
92 |
+
|
93 |
+
chain = ConversationalRetrievalChain.from_llm(
|
94 |
+
llm=llm,
|
95 |
+
retriever=retriever,
|
96 |
+
chain_type="stuff",
|
97 |
+
memory=memory,
|
98 |
+
verbose=True,
|
99 |
+
return_source_documents=True
|
100 |
+
)
|
101 |
+
return chain
|
102 |
+
|
103 |
+
# Transcribe audio using Groq Whisper
|
104 |
+
def transcribe_audio(file_path):
|
105 |
+
"""Transcribe audio using Groq's Whisper model."""
|
106 |
+
with open(file_path, "rb") as file:
|
107 |
+
transcription = groq_client.audio.transcriptions.create(
|
108 |
+
file=(file_path, file.read()),
|
109 |
+
model="distil-whisper-large-v3-en",
|
110 |
+
response_format="json",
|
111 |
+
language="en"
|
112 |
+
)
|
113 |
+
return transcription.text
|
114 |
+
|
115 |
+
# Audio Processor Class for Recording
|
116 |
+
class AudioProcessor(AudioProcessorBase):
|
117 |
+
def recv(self, frame: av.AudioFrame) -> av.AudioFrame:
|
118 |
+
return frame
|
119 |
+
|
120 |
+
# Streamlit UI
|
121 |
+
st.title("Chat with PDFs via Speech/Text ποΈππ")
|
122 |
+
|
123 |
+
uploaded_files = st.file_uploader("Upload PDF Files", accept_multiple_files=True, type=["pdf"])
|
124 |
+
|
125 |
+
if uploaded_files:
|
126 |
+
vectorstore = process_and_store_pdfs(uploaded_files)
|
127 |
+
chain = chat_chain(vectorstore)
|
128 |
+
st.success("PDFs processed! Ready to chat.")
|
129 |
+
|
130 |
+
input_method = st.radio("Choose Input Method", ["Text Input", "Record Audio", "Upload Audio File"])
|
131 |
+
|
132 |
+
# Text Input Mode
|
133 |
+
if input_method == "Text Input":
|
134 |
+
query = st.text_input("Ask your question:")
|
135 |
+
if query:
|
136 |
+
with st.spinner("Thinking..."):
|
137 |
+
response = chain({"question": query})["answer"]
|
138 |
+
st.write(f"**Response:** {response}")
|
139 |
+
|
140 |
+
# Record Audio
|
141 |
+
elif input_method == "Record Audio":
|
142 |
+
st.write("Record your audio query:")
|
143 |
+
webrtc_ctx = webrtc_streamer(
|
144 |
+
key="record",
|
145 |
+
mode=WebRtcMode.SENDONLY,
|
146 |
+
audio_receiver_size=1024,
|
147 |
+
audio_processor_factory=AudioProcessor,
|
148 |
+
media_stream_constraints={"audio": True, "video": False},
|
149 |
+
)
|
150 |
+
|
151 |
+
if webrtc_ctx.audio_receiver:
|
152 |
+
st.write("Recording...")
|
153 |
+
audio_frames = []
|
154 |
+
while True:
|
155 |
+
frame = webrtc_ctx.audio_receiver.recv()
|
156 |
+
audio_frames.append(frame)
|
157 |
+
if len(audio_frames) > 5: # Stop recording after a few frames
|
158 |
+
break
|
159 |
+
|
160 |
+
# Save the recorded audio
|
161 |
+
audio_file_path = "recorded_audio.wav"
|
162 |
+
with av.open(audio_file_path, "w") as f:
|
163 |
+
for frame in audio_frames:
|
164 |
+
f.write(frame)
|
165 |
+
st.success("Recording complete!")
|
166 |
+
|
167 |
+
# Transcribe and Generate Response
|
168 |
+
st.write("Transcribing audio...")
|
169 |
+
transcription = transcribe_audio(audio_file_path)
|
170 |
+
st.write(f"**You said:** {transcription}")
|
171 |
+
|
172 |
+
with st.spinner("Generating response..."):
|
173 |
+
response = chain({"question": transcription})["answer"]
|
174 |
+
st.write(f"**Response:** {response}")
|
175 |
+
|
176 |
+
# Upload Audio File Mode
|
177 |
+
elif input_method == "Upload Audio File":
|
178 |
+
uploaded_audio = st.file_uploader("Upload an audio file (.wav, .mp3)", type=["wav", "mp3"])
|
179 |
+
if uploaded_audio:
|
180 |
+
audio_file_path = "uploaded_audio.wav"
|
181 |
+
with open(audio_file_path, "wb") as f:
|
182 |
+
f.write(uploaded_audio.read())
|
183 |
+
|
184 |
+
st.audio(audio_file_path, format="audio/wav")
|
185 |
+
st.write("Transcribing audio...")
|
186 |
+
transcription = transcribe_audio(audio_file_path)
|
187 |
+
st.write(f"**You said:** {transcription}")
|
188 |
+
|
189 |
+
with st.spinner("Generating response..."):
|
190 |
+
response = chain({"question": transcription})["answer"]
|
191 |
+
st.write(f"**Response:** {response}")
|
192 |
+
else:
|
193 |
+
st.info("Please upload PDF files to start chatting.")
|
194 |
+
|