Spaces:
Running
Running
Update qa.py
Browse files
qa.py
CHANGED
@@ -18,9 +18,9 @@ def transcribe_audio_deepgram(local_audio_path: str) -> str:
|
|
18 |
raise ValueError("Deepgram API key not found in environment variables.")
|
19 |
|
20 |
url = "https://api.deepgram.com/v1/listen?model=nova-2&smart_format=true"
|
|
|
21 |
headers = {
|
22 |
"Authorization": f"Token {DEEPGRAM_API_KEY}",
|
23 |
-
# Adjust Content-Type if user upload might be MP3: "audio/mpeg"
|
24 |
"Content-Type": "audio/wav"
|
25 |
}
|
26 |
|
@@ -51,11 +51,10 @@ def call_llm_for_qa(conversation_so_far: str, user_question: str) -> dict:
|
|
51 |
{{ "speaker": "John", "text": "Sure, here's my answer..." }}
|
52 |
"""
|
53 |
|
54 |
-
# We'll rely on a helper in utils.py to do the Groq call:
|
55 |
from utils import call_groq_api_for_qa
|
56 |
|
57 |
raw_json_response = call_groq_api_for_qa(system_prompt)
|
58 |
-
# Expect a JSON string
|
59 |
response_dict = json.loads(raw_json_response)
|
60 |
return response_dict
|
61 |
|
@@ -69,13 +68,12 @@ def handle_qa_exchange(user_question: str) -> (bytes, str):
|
|
69 |
"""
|
70 |
conversation_so_far = st.session_state.get("conversation_history", "")
|
71 |
|
72 |
-
# Ask LLM
|
73 |
response_dict = call_llm_for_qa(conversation_so_far, user_question)
|
74 |
answer_text = response_dict.get("text", "")
|
75 |
speaker = response_dict.get("speaker", "John")
|
76 |
|
77 |
# Update conversation
|
78 |
-
# e.g., "User: question" and "John: answer"
|
79 |
new_history = conversation_so_far + f"\nUser: {user_question}\n{speaker}: {answer_text}\n"
|
80 |
st.session_state["conversation_history"] = new_history
|
81 |
|
@@ -83,7 +81,7 @@ def handle_qa_exchange(user_question: str) -> (bytes, str):
|
|
83 |
return (None, "")
|
84 |
|
85 |
# TTS
|
86 |
-
audio_file_path = generate_audio_mp3(answer_text, "John") #
|
87 |
with open(audio_file_path, "rb") as f:
|
88 |
audio_bytes = f.read()
|
89 |
|
|
|
18 |
raise ValueError("Deepgram API key not found in environment variables.")
|
19 |
|
20 |
url = "https://api.deepgram.com/v1/listen?model=nova-2&smart_format=true"
|
21 |
+
# For WAV -> "audio/wav". If user uploads MP3, you'd use "audio/mpeg".
|
22 |
headers = {
|
23 |
"Authorization": f"Token {DEEPGRAM_API_KEY}",
|
|
|
24 |
"Content-Type": "audio/wav"
|
25 |
}
|
26 |
|
|
|
51 |
{{ "speaker": "John", "text": "Sure, here's my answer..." }}
|
52 |
"""
|
53 |
|
|
|
54 |
from utils import call_groq_api_for_qa
|
55 |
|
56 |
raw_json_response = call_groq_api_for_qa(system_prompt)
|
57 |
+
# Expect a JSON string: {"speaker": "John", "text": "some short answer"}
|
58 |
response_dict = json.loads(raw_json_response)
|
59 |
return response_dict
|
60 |
|
|
|
68 |
"""
|
69 |
conversation_so_far = st.session_state.get("conversation_history", "")
|
70 |
|
71 |
+
# Ask the LLM
|
72 |
response_dict = call_llm_for_qa(conversation_so_far, user_question)
|
73 |
answer_text = response_dict.get("text", "")
|
74 |
speaker = response_dict.get("speaker", "John")
|
75 |
|
76 |
# Update conversation
|
|
|
77 |
new_history = conversation_so_far + f"\nUser: {user_question}\n{speaker}: {answer_text}\n"
|
78 |
st.session_state["conversation_history"] = new_history
|
79 |
|
|
|
81 |
return (None, "")
|
82 |
|
83 |
# TTS
|
84 |
+
audio_file_path = generate_audio_mp3(answer_text, "John") # always John
|
85 |
with open(audio_file_path, "rb") as f:
|
86 |
audio_bytes = f.read()
|
87 |
|