remove nexus and refactor stt
Browse files- kitt/core/stt.py +39 -0
- main.py +9 -195
kitt/core/stt.py
ADDED
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import copy
|
2 |
+
import time
|
3 |
+
|
4 |
+
import gradio as gr
|
5 |
+
import numpy as np
|
6 |
+
import torch
|
7 |
+
import torchaudio
|
8 |
+
from loguru import logger
|
9 |
+
from transformers import pipeline
|
10 |
+
|
11 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
12 |
+
transcriber = pipeline(
|
13 |
+
"automatic-speech-recognition", model="openai/whisper-base.en", device=device
|
14 |
+
)
|
15 |
+
|
16 |
+
|
17 |
+
def save_audio_as_wav(data, sample_rate, file_path):
|
18 |
+
# make a tensor from the numpy array
|
19 |
+
data = torch.tensor(data).reshape(1, -1)
|
20 |
+
torchaudio.save(
|
21 |
+
file_path, data, sample_rate=sample_rate, bits_per_sample=16, encoding="PCM_S"
|
22 |
+
)
|
23 |
+
|
24 |
+
|
25 |
+
def save_and_transcribe_audio(audio):
|
26 |
+
sample_rate, data = audio
|
27 |
+
try:
|
28 |
+
# add timestamp to file name
|
29 |
+
filename = f"recordings/audio{time.time()}.wav"
|
30 |
+
save_audio_as_wav(data, sample_rate, filename)
|
31 |
+
data = data.astype(np.float32)
|
32 |
+
data /= np.max(np.abs(data))
|
33 |
+
text = transcriber({"sampling_rate": sample_rate, "raw": data})["text"]
|
34 |
+
gr.Info(f"Transcribed text is: {text}\nProcessing the input...")
|
35 |
+
|
36 |
+
except Exception as e:
|
37 |
+
logger.error(f"Error: {e}")
|
38 |
+
raise Exception("Error transcribing audio.")
|
39 |
+
return text
|
main.py
CHANGED
@@ -1,24 +1,14 @@
|
|
1 |
-
import time
|
2 |
-
|
3 |
import gradio as gr
|
4 |
-
import numpy as np
|
5 |
-
import ollama
|
6 |
-
import torch
|
7 |
-
import torchaudio
|
8 |
-
import typer
|
9 |
from langchain.memory import ChatMessageHistory
|
10 |
from langchain.tools import tool
|
11 |
-
from langchain.tools.base import StructuredTool
|
12 |
from langchain_core.utils.function_calling import convert_to_openai_tool
|
13 |
from loguru import logger
|
14 |
-
from transformers import pipeline
|
15 |
|
16 |
from kitt.core import tts_gradio
|
17 |
from kitt.core import utils as kitt_utils
|
18 |
from kitt.core import voice_options
|
19 |
-
|
20 |
-
# from kitt.core.model import process_query
|
21 |
from kitt.core.model import generate_function_call as process_query
|
|
|
22 |
from kitt.core.tts import prep_for_tts, run_melo_tts, run_tts_fast, run_tts_replicate
|
23 |
from kitt.skills import (
|
24 |
code_interpreter,
|
@@ -34,7 +24,6 @@ from kitt.skills import (
|
|
34 |
set_vehicle_destination,
|
35 |
set_vehicle_speed,
|
36 |
)
|
37 |
-
from kitt.skills import vehicle_status as vehicle_status_fn
|
38 |
from kitt.skills.common import config, vehicle
|
39 |
from kitt.skills.routing import calculate_route, find_address
|
40 |
|
@@ -65,54 +54,6 @@ global_context = {
|
|
65 |
speaker_embedding_cache = {}
|
66 |
history = ChatMessageHistory()
|
67 |
|
68 |
-
MODEL_FUNC = "nexusraven"
|
69 |
-
MODEL_GENERAL = "llama3:instruct"
|
70 |
-
|
71 |
-
RAVEN_PROMPT_FUNC = """You are a helpful AI assistant in a car (vehicle), that follows instructions extremely well. \
|
72 |
-
Answer questions concisely and do not mention what you base your reply on."
|
73 |
-
|
74 |
-
{raven_tools}
|
75 |
-
|
76 |
-
{history}
|
77 |
-
|
78 |
-
User Query: Question: {input}<human_end>
|
79 |
-
"""
|
80 |
-
|
81 |
-
|
82 |
-
HERMES_PROMPT_FUNC = """
|
83 |
-
<|im_start|>system
|
84 |
-
You are a helpful AI assistant in a car (vehicle), that follows instructions extremely well. \
|
85 |
-
Answer questions concisely and do not mention what you base your reply on.<|im_end|>
|
86 |
-
<|im_start|>user
|
87 |
-
{{ .Prompt }}<|im_end|>
|
88 |
-
<|im_start|>assistant
|
89 |
-
"""
|
90 |
-
|
91 |
-
|
92 |
-
def get_prompt(template, input, history, tools):
|
93 |
-
# "vehicle_status": vehicle_status_fn()[0]
|
94 |
-
kwargs = {"history": history, "input": input}
|
95 |
-
prompt = "<human>:\n"
|
96 |
-
for tool in tools:
|
97 |
-
func_signature, func_docstring = tool.description.split(" - ", 1)
|
98 |
-
prompt += f'Function:\n<func_start>def {func_signature}<func_end>\n<docstring_start>\n"""\n{func_docstring}\n"""\n<docstring_end>\n'
|
99 |
-
kwargs["raven_tools"] = prompt
|
100 |
-
|
101 |
-
if history:
|
102 |
-
kwargs["history"] = f"Previous conversation history:{history}\n"
|
103 |
-
|
104 |
-
return template.format(**kwargs).replace("{{", "{").replace("}}", "}")
|
105 |
-
|
106 |
-
|
107 |
-
def use_tool(func_name, kwargs, tools):
|
108 |
-
for tool in tools:
|
109 |
-
if tool.name == func_name:
|
110 |
-
return tool.invoke(input=kwargs)
|
111 |
-
return None
|
112 |
-
|
113 |
-
|
114 |
-
# llm = Ollama(model="nexusraven", stop=["\nReflection:", "\nThought:"], keep_alive=60*10)
|
115 |
-
|
116 |
|
117 |
# Generate options for hours (00-23)
|
118 |
hour_options = [f"{i:02d}:00:00" for i in range(24)]
|
@@ -136,20 +77,6 @@ def set_time(time_picker):
|
|
136 |
return vehicle
|
137 |
|
138 |
|
139 |
-
tools = [
|
140 |
-
# StructuredTool.from_function(get_weather),
|
141 |
-
# StructuredTool.from_function(find_route),
|
142 |
-
# StructuredTool.from_function(vehicle_status_fn),
|
143 |
-
# StructuredTool.from_function(set_vehicle_speed),
|
144 |
-
# StructuredTool.from_function(set_vehicle_destination),
|
145 |
-
# StructuredTool.from_function(search_points_of_interest),
|
146 |
-
# StructuredTool.from_function(search_along_route),
|
147 |
-
# StructuredTool.from_function(date_time_info),
|
148 |
-
# StructuredTool.from_function(get_weather_current_location),
|
149 |
-
# StructuredTool.from_function(code_interpreter),
|
150 |
-
# StructuredTool.from_function(do_anything_else),
|
151 |
-
]
|
152 |
-
|
153 |
functions = [
|
154 |
# set_vehicle_speed,
|
155 |
set_vehicle_destination,
|
@@ -161,59 +88,11 @@ functions = [
|
|
161 |
openai_tools = [convert_to_openai_tool(tool) for tool in functions]
|
162 |
|
163 |
|
164 |
-
def run_generic_model(query):
|
165 |
-
print(f"Running the generic model with query: {query}")
|
166 |
-
data = {
|
167 |
-
"prompt": f"Answer the question below in a short and concise manner.\n{query}",
|
168 |
-
"model": MODEL_GENERAL,
|
169 |
-
"options": {
|
170 |
-
# "temperature": 0.1,
|
171 |
-
# "stop":["\nReflection:", "\nThought:"]
|
172 |
-
},
|
173 |
-
}
|
174 |
-
out = ollama.generate(**data)
|
175 |
-
return out["response"]
|
176 |
-
|
177 |
-
|
178 |
def clear_history():
|
179 |
logger.info("Clearing the conversation history...")
|
180 |
history.clear()
|
181 |
|
182 |
|
183 |
-
def run_nexusraven_model(query, voice_character, state):
|
184 |
-
global_context["prompt"] = get_prompt(RAVEN_PROMPT_FUNC, query, "", tools)
|
185 |
-
print("Prompt: ", global_context["prompt"])
|
186 |
-
data = {
|
187 |
-
"prompt": global_context["prompt"],
|
188 |
-
# "streaming": False,
|
189 |
-
"model": "nexusraven",
|
190 |
-
# "model": "smangrul/llama-3-8b-instruct-function-calling",
|
191 |
-
"raw": True,
|
192 |
-
"options": {"temperature": 0.5, "stop": ["\nReflection:", "\nThought:"]},
|
193 |
-
}
|
194 |
-
out = ollama.generate(**data)
|
195 |
-
llm_response = out["response"]
|
196 |
-
if "Call: " in llm_response:
|
197 |
-
print(f"llm_response: {llm_response}")
|
198 |
-
llm_response = llm_response.replace("<bot_end>", " ")
|
199 |
-
func_name, kwargs = extract_func_args(llm_response)
|
200 |
-
print(f"Function: {func_name}, Args: {kwargs}")
|
201 |
-
if func_name == "do_anything_else":
|
202 |
-
output_text = run_generic_model(query)
|
203 |
-
else:
|
204 |
-
output_text = use_tool(func_name, kwargs, tools)
|
205 |
-
else:
|
206 |
-
output_text = out["response"]
|
207 |
-
|
208 |
-
if type(output_text) == tuple:
|
209 |
-
output_text = output_text[0]
|
210 |
-
gr.Info(f"Output text: {output_text}\nGenerating voice output...")
|
211 |
-
return (
|
212 |
-
output_text,
|
213 |
-
tts_gradio(output_text, voice_character, speaker_embedding_cache)[0],
|
214 |
-
)
|
215 |
-
|
216 |
-
|
217 |
def run_llama3_model(query, voice_character, state):
|
218 |
|
219 |
assert len(functions) > 0, "No functions to call"
|
@@ -249,18 +128,13 @@ def run_llama3_model(query, voice_character, state):
|
|
249 |
|
250 |
|
251 |
def run_model(query, voice_character, state):
|
252 |
-
model = state.get("model", "
|
253 |
query = query.strip().replace("'", "")
|
254 |
logger.info(
|
255 |
f"Running model: {model} with query: {query}, voice_character: {voice_character} and llm_backend: {state['llm_backend']}, tts_enabled: {state['tts_enabled']}"
|
256 |
)
|
257 |
global_context["query"] = query
|
258 |
-
|
259 |
-
text, voice = run_nexusraven_model(query, voice_character, state)
|
260 |
-
elif model == "llama3":
|
261 |
-
text, voice = run_llama3_model(query, voice_character, state)
|
262 |
-
else:
|
263 |
-
text, voice = "Error running model", None
|
264 |
|
265 |
if not state["enable_history"]:
|
266 |
history.clear()
|
@@ -308,44 +182,6 @@ def update_vehicle_status(trip_progress, origin, destination, state):
|
|
308 |
return vehicle, plot, state
|
309 |
|
310 |
|
311 |
-
device = "cuda" if torch.cuda.is_available() else "cpu"
|
312 |
-
transcriber = pipeline(
|
313 |
-
"automatic-speech-recognition", model="openai/whisper-base.en", device=device
|
314 |
-
)
|
315 |
-
|
316 |
-
|
317 |
-
def save_audio_as_wav(data, sample_rate, file_path):
|
318 |
-
# make a tensor from the numpy array
|
319 |
-
data = torch.tensor(data).reshape(1, -1)
|
320 |
-
torchaudio.save(
|
321 |
-
file_path, data, sample_rate=sample_rate, bits_per_sample=16, encoding="PCM_S"
|
322 |
-
)
|
323 |
-
|
324 |
-
|
325 |
-
def save_and_transcribe_audio(audio):
|
326 |
-
try:
|
327 |
-
# capture the audio and save it to a file as wav or mp3
|
328 |
-
# file_name = save("audioinput.wav")
|
329 |
-
sr, y = audio
|
330 |
-
# y = y.astype(np.float32)
|
331 |
-
# y /= np.max(np.abs(y))
|
332 |
-
|
333 |
-
# add timestamp to file name
|
334 |
-
filename = f"recordings/audio{time.time()}.wav"
|
335 |
-
save_audio_as_wav(y, sr, filename)
|
336 |
-
|
337 |
-
sr, y = audio
|
338 |
-
y = y.astype(np.float32)
|
339 |
-
y /= np.max(np.abs(y))
|
340 |
-
text = transcriber({"sampling_rate": sr, "raw": y})["text"]
|
341 |
-
gr.Info(f"Transcribed text is: {text}\nProcessing the input...")
|
342 |
-
|
343 |
-
except Exception as e:
|
344 |
-
logger.error(f"Error: {e}")
|
345 |
-
raise Exception("Error transcribing audio.")
|
346 |
-
return text
|
347 |
-
|
348 |
-
|
349 |
def save_and_transcribe_run_model(audio, voice_character, state):
|
350 |
text = save_and_transcribe_audio(audio)
|
351 |
out_text, out_voice, vehicle_status, state, update_proxy = run_model(
|
@@ -494,7 +330,12 @@ def create_demo(tts_server: bool = False, model="llama3"):
|
|
494 |
0, 100, step=5, label="Trip progress", interactive=True
|
495 |
)
|
496 |
|
497 |
-
|
|
|
|
|
|
|
|
|
|
|
498 |
|
499 |
with gr.Row():
|
500 |
with gr.Column():
|
@@ -647,30 +488,3 @@ demo.launch(
|
|
647 |
ssl_verify=False,
|
648 |
share=False,
|
649 |
)
|
650 |
-
|
651 |
-
app = typer.Typer()
|
652 |
-
|
653 |
-
|
654 |
-
@app.command()
|
655 |
-
def run(tts_server: bool = False):
|
656 |
-
global demo
|
657 |
-
demo = create_demo(tts_server)
|
658 |
-
demo.launch(
|
659 |
-
debug=True, server_name="0.0.0.0", server_port=7860, ssl_verify=True, share=True
|
660 |
-
)
|
661 |
-
|
662 |
-
|
663 |
-
@app.command()
|
664 |
-
def dev(tts_server: bool = False, model: str = "llama3"):
|
665 |
-
demo = create_demo(tts_server, model)
|
666 |
-
demo.launch(
|
667 |
-
debug=True,
|
668 |
-
server_name="0.0.0.0",
|
669 |
-
server_port=7860,
|
670 |
-
ssl_verify=False,
|
671 |
-
share=False,
|
672 |
-
)
|
673 |
-
|
674 |
-
|
675 |
-
if __name__ == "__main__":
|
676 |
-
app()
|
|
|
|
|
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
2 |
from langchain.memory import ChatMessageHistory
|
3 |
from langchain.tools import tool
|
|
|
4 |
from langchain_core.utils.function_calling import convert_to_openai_tool
|
5 |
from loguru import logger
|
|
|
6 |
|
7 |
from kitt.core import tts_gradio
|
8 |
from kitt.core import utils as kitt_utils
|
9 |
from kitt.core import voice_options
|
|
|
|
|
10 |
from kitt.core.model import generate_function_call as process_query
|
11 |
+
from kitt.core.stt import save_and_transcribe_audio
|
12 |
from kitt.core.tts import prep_for_tts, run_melo_tts, run_tts_fast, run_tts_replicate
|
13 |
from kitt.skills import (
|
14 |
code_interpreter,
|
|
|
24 |
set_vehicle_destination,
|
25 |
set_vehicle_speed,
|
26 |
)
|
|
|
27 |
from kitt.skills.common import config, vehicle
|
28 |
from kitt.skills.routing import calculate_route, find_address
|
29 |
|
|
|
54 |
speaker_embedding_cache = {}
|
55 |
history = ChatMessageHistory()
|
56 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
|
58 |
# Generate options for hours (00-23)
|
59 |
hour_options = [f"{i:02d}:00:00" for i in range(24)]
|
|
|
77 |
return vehicle
|
78 |
|
79 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
80 |
functions = [
|
81 |
# set_vehicle_speed,
|
82 |
set_vehicle_destination,
|
|
|
88 |
openai_tools = [convert_to_openai_tool(tool) for tool in functions]
|
89 |
|
90 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
def clear_history():
|
92 |
logger.info("Clearing the conversation history...")
|
93 |
history.clear()
|
94 |
|
95 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
96 |
def run_llama3_model(query, voice_character, state):
|
97 |
|
98 |
assert len(functions) > 0, "No functions to call"
|
|
|
128 |
|
129 |
|
130 |
def run_model(query, voice_character, state):
|
131 |
+
model = state.get("model", "llama3")
|
132 |
query = query.strip().replace("'", "")
|
133 |
logger.info(
|
134 |
f"Running model: {model} with query: {query}, voice_character: {voice_character} and llm_backend: {state['llm_backend']}, tts_enabled: {state['tts_enabled']}"
|
135 |
)
|
136 |
global_context["query"] = query
|
137 |
+
text, voice = run_llama3_model(query, voice_character, state)
|
|
|
|
|
|
|
|
|
|
|
138 |
|
139 |
if not state["enable_history"]:
|
140 |
history.clear()
|
|
|
182 |
return vehicle, plot, state
|
183 |
|
184 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
185 |
def save_and_transcribe_run_model(audio, voice_character, state):
|
186 |
text = save_and_transcribe_audio(audio)
|
187 |
out_text, out_voice, vehicle_status, state, update_proxy = run_model(
|
|
|
330 |
0, 100, step=5, label="Trip progress", interactive=True
|
331 |
)
|
332 |
|
333 |
+
# with gr.Column(scale=1, min_width=300):
|
334 |
+
# gr.Image("linkedin-1.png", label="Linkedin - Sasan Jafarnejad")
|
335 |
+
# gr.Image(
|
336 |
+
# "team-ubix.png",
|
337 |
+
# label="Research Team - UBIX - University of Luxembourg",
|
338 |
+
# )
|
339 |
|
340 |
with gr.Row():
|
341 |
with gr.Column():
|
|
|
488 |
ssl_verify=False,
|
489 |
share=False,
|
490 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|