Spaces:
Running
Running
Update qa.py
Browse files
qa.py
CHANGED
@@ -1,40 +1,33 @@
|
|
1 |
# qa.py
|
2 |
|
3 |
import os
|
|
|
4 |
import json
|
5 |
import tempfile
|
6 |
import streamlit as st
|
7 |
-
from streamlit_webrtc import webrtc_streamer, WebRtcMode, RTCConfiguration, AudioProcessorBase
|
8 |
|
9 |
from utils import generate_audio_mp3, call_groq_api_for_qa
|
10 |
-
|
11 |
import av
|
12 |
import pydub
|
13 |
-
import wave
|
14 |
import numpy as np
|
15 |
|
16 |
-
|
17 |
-
RTC_CONFIGURATION = RTCConfiguration(
|
18 |
-
{"iceServers": [{"urls": ["stun:stun.l.google.com:19302"]}]}
|
19 |
-
)
|
20 |
|
21 |
class AudioBufferProcessor(AudioProcessorBase):
|
22 |
"""
|
23 |
A custom audio processor that accumulates raw audio frames in memory.
|
24 |
-
|
25 |
"""
|
26 |
def __init__(self) -> None:
|
27 |
self.frames = []
|
|
|
28 |
|
29 |
def recv_audio(self, frame: av.AudioFrame) -> av.AudioFrame:
|
30 |
-
|
|
|
31 |
pcm = frame.to_ndarray()
|
32 |
-
# The shape is (channels, samples)
|
33 |
-
# We'll assume single channel or handle the first channel
|
34 |
if pcm.ndim == 2 and pcm.shape[0] > 1:
|
35 |
-
# If stereo, just take the first channel for STT
|
36 |
pcm = pcm[0, :]
|
37 |
-
|
38 |
sample_rate = frame.sample_rate
|
39 |
samples = pcm.astype(np.int16).tobytes()
|
40 |
segment = pydub.AudioSegment(
|
@@ -48,8 +41,7 @@ class AudioBufferProcessor(AudioProcessorBase):
|
|
48 |
|
49 |
def finalize_wav(self) -> str:
|
50 |
"""
|
51 |
-
|
52 |
-
Returns path to the wav file.
|
53 |
"""
|
54 |
if not self.frames:
|
55 |
return ""
|
@@ -58,6 +50,25 @@ class AudioBufferProcessor(AudioProcessorBase):
|
|
58 |
combined.export(tmp_wav.name, format="wav")
|
59 |
return tmp_wav.name
|
60 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
|
62 |
def handle_qa_exchange(conversation_so_far: str, user_question: str) -> (bytes, str):
|
63 |
"""
|
@@ -86,7 +97,6 @@ def handle_qa_exchange(conversation_so_far: str, user_question: str) -> (bytes,
|
|
86 |
if not answer_text.strip():
|
87 |
return (None, "")
|
88 |
|
89 |
-
# TTS
|
90 |
audio_file_path = generate_audio_mp3(answer_text, "John")
|
91 |
with open(audio_file_path, "rb") as f:
|
92 |
audio_bytes = f.read()
|
|
|
1 |
# qa.py
|
2 |
|
3 |
import os
|
4 |
+
import requests
|
5 |
import json
|
6 |
import tempfile
|
7 |
import streamlit as st
|
|
|
8 |
|
9 |
from utils import generate_audio_mp3, call_groq_api_for_qa
|
|
|
10 |
import av
|
11 |
import pydub
|
|
|
12 |
import numpy as np
|
13 |
|
14 |
+
from streamlit_webrtc import AudioProcessorBase
|
|
|
|
|
|
|
15 |
|
16 |
class AudioBufferProcessor(AudioProcessorBase):
|
17 |
"""
|
18 |
A custom audio processor that accumulates raw audio frames in memory.
|
19 |
+
Now we also count frames for debugging.
|
20 |
"""
|
21 |
def __init__(self) -> None:
|
22 |
self.frames = []
|
23 |
+
self.frame_count = 0
|
24 |
|
25 |
def recv_audio(self, frame: av.AudioFrame) -> av.AudioFrame:
|
26 |
+
self.frame_count += 1 # debug increment
|
27 |
+
|
28 |
pcm = frame.to_ndarray()
|
|
|
|
|
29 |
if pcm.ndim == 2 and pcm.shape[0] > 1:
|
|
|
30 |
pcm = pcm[0, :]
|
|
|
31 |
sample_rate = frame.sample_rate
|
32 |
samples = pcm.astype(np.int16).tobytes()
|
33 |
segment = pydub.AudioSegment(
|
|
|
41 |
|
42 |
def finalize_wav(self) -> str:
|
43 |
"""
|
44 |
+
Combine frames into one WAV file if we have them.
|
|
|
45 |
"""
|
46 |
if not self.frames:
|
47 |
return ""
|
|
|
50 |
combined.export(tmp_wav.name, format="wav")
|
51 |
return tmp_wav.name
|
52 |
|
53 |
+
def transcribe_audio_deepgram(local_audio_path: str) -> str:
|
54 |
+
"""
|
55 |
+
Sends a local audio file to Deepgram STT, returns the transcript text if successful.
|
56 |
+
"""
|
57 |
+
DEEPGRAM_API_KEY = os.environ.get("DEEPGRAM_API_KEY")
|
58 |
+
if not DEEPGRAM_API_KEY:
|
59 |
+
raise ValueError("Deepgram API key not found in environment variables.")
|
60 |
+
|
61 |
+
url = "https://api.deepgram.com/v1/listen?model=nova-2&smart_format=true"
|
62 |
+
headers = {
|
63 |
+
"Authorization": f"Token {DEEPGRAM_API_KEY}",
|
64 |
+
"Content-Type": "audio/wav"
|
65 |
+
}
|
66 |
+
with open(local_audio_path, "rb") as f:
|
67 |
+
response = requests.post(url, headers=headers, data=f)
|
68 |
+
response.raise_for_status()
|
69 |
+
data = response.json()
|
70 |
+
transcript = data["results"]["channels"][0]["alternatives"][0].get("transcript", "")
|
71 |
+
return transcript
|
72 |
|
73 |
def handle_qa_exchange(conversation_so_far: str, user_question: str) -> (bytes, str):
|
74 |
"""
|
|
|
97 |
if not answer_text.strip():
|
98 |
return (None, "")
|
99 |
|
|
|
100 |
audio_file_path = generate_audio_mp3(answer_text, "John")
|
101 |
with open(audio_file_path, "rb") as f:
|
102 |
audio_bytes = f.read()
|