Commit
·
0af1295
1
Parent(s):
e0a450a
add app
Browse files- assistant-streamlit.py +51 -0
assistant-streamlit.py
ADDED
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import sounddevice as sd
|
3 |
+
import soundfile as sf
|
4 |
+
from faster_whisper import WhisperModel
|
5 |
+
import io
|
6 |
+
import os
|
7 |
+
from langchain_community.llms import Ollama
|
8 |
+
import pyttsx3
|
9 |
+
# Set environment variable to handle duplicate libraries
|
10 |
+
os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
|
11 |
+
|
12 |
+
# Initialize WhisperModel and Ollama
|
13 |
+
model_size = "base.en"
|
14 |
+
model = WhisperModel(model_size, device="cpu", compute_type="int8", num_workers=5)
|
15 |
+
llm = Ollama(model="tinyllama")
|
16 |
+
|
17 |
+
# Initialize text-to-speech engine
|
18 |
+
engine = pyttsx3.init('sapi5')
|
19 |
+
voices = engine.getProperty('voices')
|
20 |
+
engine.setProperty('voice',voices[0].id)
|
21 |
+
engine.setProperty('rate',180)
|
22 |
+
|
23 |
+
def speak(audio):
|
24 |
+
engine.say(audio)
|
25 |
+
engine.runAndWait()
|
26 |
+
|
27 |
+
# Record and transcribe audio
|
28 |
+
audio_data = st.audio("recorded_audio.wav", format="audio/wav", start_time=0)
|
29 |
+
if st.button("Record"):
|
30 |
+
with st.spinner("Recording..."):
|
31 |
+
recorded_audio = sd.rec(int(5 * 44100), samplerate=44100, channels=2, dtype="int16")
|
32 |
+
sd.wait()
|
33 |
+
sf.write("recorded_audio.wav", recorded_audio, samplerate=44100)
|
34 |
+
|
35 |
+
st.audio("recorded_audio.wav", format="audio/wav", start_time=0)
|
36 |
+
|
37 |
+
# Transcribe audio and speak response
|
38 |
+
with open("recorded_audio.wav", "rb") as audio_file:
|
39 |
+
segments,info= model.transcribe(io.BytesIO(audio_file.read()), beam_size=10)
|
40 |
+
for segment in segments:
|
41 |
+
prompt=segment.text
|
42 |
+
print(prompt)
|
43 |
+
st.text(prompt)
|
44 |
+
if prompt:
|
45 |
+
response = llm.invoke(prompt)
|
46 |
+
st.success("Response: " + response)
|
47 |
+
speak(response)
|
48 |
+
st.stop()
|
49 |
+
else:
|
50 |
+
st.error("Failed to transcribe audio.")
|
51 |
+
|