Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,27 +1,30 @@
|
|
1 |
import streamlit as st
|
2 |
import os
|
|
|
|
|
3 |
from pydub import AudioSegment
|
4 |
from pydub.silence import split_on_silence
|
5 |
from dotenv import load_dotenv
|
6 |
from tempfile import NamedTemporaryFile
|
7 |
import math
|
8 |
from docx import Document
|
9 |
-
import
|
10 |
|
11 |
-
# Load environment variables from .env file (if needed
|
12 |
load_dotenv()
|
13 |
|
14 |
@st.cache_resource
|
15 |
def load_whisper_model():
|
16 |
"""
|
17 |
-
Load the Whisper model
|
18 |
-
You can
|
19 |
"""
|
20 |
-
|
21 |
-
|
|
|
|
|
22 |
|
23 |
-
|
24 |
-
model = load_whisper_model()
|
25 |
|
26 |
def split_audio_on_silence(audio_file_path, min_silence_len=500, silence_thresh=-40, keep_silence=250):
|
27 |
"""
|
@@ -47,7 +50,7 @@ def split_audio_on_silence(audio_file_path, min_silence_len=500, silence_thresh=
|
|
47 |
|
48 |
def transcribe(audio_file):
|
49 |
"""
|
50 |
-
Transcribe an audio file using the locally loaded Whisper model.
|
51 |
|
52 |
Args:
|
53 |
audio_file (str): Path to the audio file.
|
@@ -55,12 +58,17 @@ def transcribe(audio_file):
|
|
55 |
Returns:
|
56 |
str: Transcribed text.
|
57 |
"""
|
58 |
-
|
59 |
-
|
|
|
|
|
|
|
|
|
|
|
60 |
|
61 |
def process_audio_chunks(audio_chunks):
|
62 |
"""
|
63 |
-
Process and transcribe each audio chunk
|
64 |
|
65 |
Args:
|
66 |
audio_chunks (list): List of AudioSegment chunks.
|
@@ -69,22 +77,22 @@ def process_audio_chunks(audio_chunks):
|
|
69 |
str: Combined transcription from all chunks.
|
70 |
"""
|
71 |
transcriptions = []
|
72 |
-
min_length_ms = 100 # Minimum length required
|
|
|
73 |
for i, chunk in enumerate(audio_chunks):
|
74 |
if len(chunk) < min_length_ms:
|
75 |
st.warning(f"Chunk {i} is too short to be processed.")
|
76 |
continue
|
77 |
|
78 |
-
# Save the chunk temporarily as a WAV file
|
79 |
with NamedTemporaryFile(delete=False, suffix=".wav") as temp_audio_file:
|
80 |
chunk.export(temp_audio_file.name, format="wav")
|
81 |
temp_audio_file_path = temp_audio_file.name
|
82 |
-
|
83 |
transcription = transcribe(temp_audio_file_path)
|
84 |
if transcription:
|
85 |
transcriptions.append(transcription)
|
86 |
st.write(f"Transcription for chunk {i}: {transcription}")
|
87 |
-
|
88 |
os.remove(temp_audio_file_path)
|
89 |
return " ".join(transcriptions)
|
90 |
|
@@ -106,7 +114,7 @@ def save_transcription_to_docx(transcription, audio_file_path):
|
|
106 |
doc.save(output_file_name)
|
107 |
return output_file_name
|
108 |
|
109 |
-
st.title("Audio Transcription with Whisper (Local)")
|
110 |
|
111 |
# Allow uploading of audio or video files
|
112 |
uploaded_file = st.file_uploader("Upload an audio or video file", type=["wav", "mp3", "ogg", "m4a", "mp4", "mov"])
|
@@ -122,14 +130,14 @@ if uploaded_file is not None and st.session_state.transcription is None:
|
|
122 |
temp_audio_file = f"temp_audio_file.{file_extension}"
|
123 |
with open(temp_audio_file, "wb") as f:
|
124 |
f.write(uploaded_file.getbuffer())
|
125 |
-
|
126 |
# Split and process audio using silence detection
|
127 |
with st.spinner('Transcribing...'):
|
128 |
audio_chunks = split_audio_on_silence(
|
129 |
temp_audio_file,
|
130 |
-
min_silence_len=500,
|
131 |
-
silence_thresh=-40,
|
132 |
-
keep_silence=250
|
133 |
)
|
134 |
transcription = process_audio_chunks(audio_chunks)
|
135 |
if transcription:
|
@@ -137,7 +145,7 @@ if uploaded_file is not None and st.session_state.transcription is None:
|
|
137 |
st.success('Transcription complete!')
|
138 |
output_docx_file = save_transcription_to_docx(transcription, uploaded_file.name)
|
139 |
st.session_state.output_docx_file = output_docx_file
|
140 |
-
|
141 |
if os.path.exists(temp_audio_file):
|
142 |
os.remove(temp_audio_file)
|
143 |
|
|
|
1 |
import streamlit as st
|
2 |
import os
|
3 |
+
import librosa
|
4 |
+
import torch
|
5 |
from pydub import AudioSegment
|
6 |
from pydub.silence import split_on_silence
|
7 |
from dotenv import load_dotenv
|
8 |
from tempfile import NamedTemporaryFile
|
9 |
import math
|
10 |
from docx import Document
|
11 |
+
from transformers import WhisperProcessor, WhisperForConditionalGeneration
|
12 |
|
13 |
+
# Load environment variables from .env file (if needed)
|
14 |
load_dotenv()
|
15 |
|
16 |
@st.cache_resource
|
17 |
def load_whisper_model():
|
18 |
"""
|
19 |
+
Load the Whisper model and processor from Hugging Face.
|
20 |
+
You can change the model variant ("openai/whisper-base" is used here).
|
21 |
"""
|
22 |
+
model_name = "openai/whisper-base" # Options: "tiny", "base", "small", "medium", "large"
|
23 |
+
processor = WhisperProcessor.from_pretrained(model_name)
|
24 |
+
model = WhisperForConditionalGeneration.from_pretrained(model_name)
|
25 |
+
return processor, model
|
26 |
|
27 |
+
processor, model = load_whisper_model()
|
|
|
28 |
|
29 |
def split_audio_on_silence(audio_file_path, min_silence_len=500, silence_thresh=-40, keep_silence=250):
|
30 |
"""
|
|
|
50 |
|
51 |
def transcribe(audio_file):
|
52 |
"""
|
53 |
+
Transcribe an audio file using the locally loaded Whisper model from Hugging Face.
|
54 |
|
55 |
Args:
|
56 |
audio_file (str): Path to the audio file.
|
|
|
58 |
Returns:
|
59 |
str: Transcribed text.
|
60 |
"""
|
61 |
+
# Load audio using librosa, resampling to 16000 Hz as required by Whisper
|
62 |
+
speech, sr = librosa.load(audio_file, sr=16000)
|
63 |
+
input_features = processor(speech, sampling_rate=16000, return_tensors="pt").input_features
|
64 |
+
# Generate transcription
|
65 |
+
predicted_ids = model.generate(input_features)
|
66 |
+
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
|
67 |
+
return transcription
|
68 |
|
69 |
def process_audio_chunks(audio_chunks):
|
70 |
"""
|
71 |
+
Process and transcribe each audio chunk.
|
72 |
|
73 |
Args:
|
74 |
audio_chunks (list): List of AudioSegment chunks.
|
|
|
77 |
str: Combined transcription from all chunks.
|
78 |
"""
|
79 |
transcriptions = []
|
80 |
+
min_length_ms = 100 # Minimum length required (0.1 seconds)
|
81 |
+
|
82 |
for i, chunk in enumerate(audio_chunks):
|
83 |
if len(chunk) < min_length_ms:
|
84 |
st.warning(f"Chunk {i} is too short to be processed.")
|
85 |
continue
|
86 |
|
|
|
87 |
with NamedTemporaryFile(delete=False, suffix=".wav") as temp_audio_file:
|
88 |
chunk.export(temp_audio_file.name, format="wav")
|
89 |
temp_audio_file_path = temp_audio_file.name
|
90 |
+
|
91 |
transcription = transcribe(temp_audio_file_path)
|
92 |
if transcription:
|
93 |
transcriptions.append(transcription)
|
94 |
st.write(f"Transcription for chunk {i}: {transcription}")
|
95 |
+
|
96 |
os.remove(temp_audio_file_path)
|
97 |
return " ".join(transcriptions)
|
98 |
|
|
|
114 |
doc.save(output_file_name)
|
115 |
return output_file_name
|
116 |
|
117 |
+
st.title("Audio Transcription with Whisper (Local via Hugging Face)")
|
118 |
|
119 |
# Allow uploading of audio or video files
|
120 |
uploaded_file = st.file_uploader("Upload an audio or video file", type=["wav", "mp3", "ogg", "m4a", "mp4", "mov"])
|
|
|
130 |
temp_audio_file = f"temp_audio_file.{file_extension}"
|
131 |
with open(temp_audio_file, "wb") as f:
|
132 |
f.write(uploaded_file.getbuffer())
|
133 |
+
|
134 |
# Split and process audio using silence detection
|
135 |
with st.spinner('Transcribing...'):
|
136 |
audio_chunks = split_audio_on_silence(
|
137 |
temp_audio_file,
|
138 |
+
min_silence_len=500,
|
139 |
+
silence_thresh=-40,
|
140 |
+
keep_silence=250
|
141 |
)
|
142 |
transcription = process_audio_chunks(audio_chunks)
|
143 |
if transcription:
|
|
|
145 |
st.success('Transcription complete!')
|
146 |
output_docx_file = save_transcription_to_docx(transcription, uploaded_file.name)
|
147 |
st.session_state.output_docx_file = output_docx_file
|
148 |
+
|
149 |
if os.path.exists(temp_audio_file):
|
150 |
os.remove(temp_audio_file)
|
151 |
|