Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,286 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import os
|
3 |
+
import numpy as np
|
4 |
+
import tempfile
|
5 |
+
import base64
|
6 |
+
import gc
|
7 |
+
import sys
|
8 |
+
import traceback
|
9 |
+
import gradio as gr
|
10 |
+
import librosa
|
11 |
+
from scipy.io.wavfile import write
|
12 |
+
from gtts import gTTS
|
13 |
+
import soundfile as sf
|
14 |
+
import whisper # Official OpenAI Whisper package
|
15 |
+
|
16 |
+
# Define device for processing
|
17 |
+
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
18 |
+
print(f"Using device: {DEVICE}")
|
19 |
+
|
20 |
+
# Free up memory
|
21 |
+
gc.collect()
|
22 |
+
if DEVICE == "cuda":
|
23 |
+
torch.cuda.empty_cache()
|
24 |
+
print(f"CUDA memory allocated: {torch.cuda.memory_allocated()/1024**2:.2f} MB")
|
25 |
+
print(f"CUDA memory reserved: {torch.cuda.memory_reserved()/1024**2:.2f} MB")
|
26 |
+
|
27 |
+
# Try importing transformers, with fallback
|
28 |
+
try:
|
29 |
+
from transformers import WhisperProcessor, WhisperForConditionalGeneration
|
30 |
+
from transformers import BertForSequenceClassification, BertTokenizer, pipeline
|
31 |
+
TRANSFORMERS_AVAILABLE = True
|
32 |
+
print("Transformers package loaded successfully")
|
33 |
+
except Exception as e:
|
34 |
+
TRANSFORMERS_AVAILABLE = False
|
35 |
+
print(f"Warning: Could not import from transformers: {e}")
|
36 |
+
|
37 |
+
class WhisperTranscriber:
|
38 |
+
def __init__(self, model_size="tiny"):
|
39 |
+
print(f"Initializing Whisper transcriber with model size: {model_size}")
|
40 |
+
self.model_size = model_size
|
41 |
+
self.processor = None
|
42 |
+
self.model = None
|
43 |
+
self.official_model = None
|
44 |
+
|
45 |
+
# Try to initialize using transformers first
|
46 |
+
if TRANSFORMERS_AVAILABLE:
|
47 |
+
try:
|
48 |
+
print(f"Loading Whisper processor: openai/whisper-{model_size}")
|
49 |
+
self.processor = WhisperProcessor.from_pretrained(f"openai/whisper-{model_size}")
|
50 |
+
|
51 |
+
print(f"Loading Whisper model: openai/whisper-{model_size}")
|
52 |
+
self.model = WhisperForConditionalGeneration.from_pretrained(f"openai/whisper-{model_size}")
|
53 |
+
|
54 |
+
if DEVICE == "cuda":
|
55 |
+
print("Moving model to CUDA")
|
56 |
+
self.model = self.model.to(DEVICE)
|
57 |
+
|
58 |
+
print("Transformers Whisper initialization complete")
|
59 |
+
except Exception as e:
|
60 |
+
print(f"Error initializing Whisper with transformers: {e}")
|
61 |
+
traceback.print_exc()
|
62 |
+
self.processor = None
|
63 |
+
self.model = None
|
64 |
+
|
65 |
+
# If transformers failed or not available, try official OpenAI implementation
|
66 |
+
if self.processor is None or self.model is None:
|
67 |
+
try:
|
68 |
+
print(f"Falling back to official OpenAI Whisper implementation with model size: {model_size}")
|
69 |
+
self.official_model = whisper.load_model(model_size)
|
70 |
+
print("Official Whisper model loaded successfully")
|
71 |
+
except Exception as e:
|
72 |
+
print(f"Error initializing official Whisper model: {e}")
|
73 |
+
traceback.print_exc()
|
74 |
+
self.official_model = None
|
75 |
+
|
76 |
+
# Check if any model was loaded
|
77 |
+
if (self.processor is None or self.model is None) and self.official_model is None:
|
78 |
+
print("WARNING: All Whisper initialization attempts failed!")
|
79 |
+
else:
|
80 |
+
print("Whisper initialized successfully with at least one implementation")
|
81 |
+
|
82 |
+
def transcribe(self, audio_path):
|
83 |
+
# Try transcribing with transformers implementation first
|
84 |
+
if self.processor is not None and self.model is not None:
|
85 |
+
try:
|
86 |
+
print("Transcribing with transformers implementation...")
|
87 |
+
|
88 |
+
# Load audio
|
89 |
+
waveform, sample_rate = librosa.load(audio_path, sr=16000)
|
90 |
+
|
91 |
+
# Process audio
|
92 |
+
input_features = self.processor(waveform, sampling_rate=16000, return_tensors="pt").input_features
|
93 |
+
if DEVICE == "cuda":
|
94 |
+
input_features = input_features.to(DEVICE)
|
95 |
+
|
96 |
+
# Generate transcription
|
97 |
+
with torch.no_grad():
|
98 |
+
predicted_ids = self.model.generate(input_features, max_length=100)
|
99 |
+
|
100 |
+
# Decode the transcription
|
101 |
+
transcription = self.processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
|
102 |
+
print("Transcription successful with transformers implementation")
|
103 |
+
return transcription
|
104 |
+
|
105 |
+
except Exception as e:
|
106 |
+
print(f"Error in transformers transcription: {e}")
|
107 |
+
traceback.print_exc()
|
108 |
+
|
109 |
+
# Fall back to official implementation if available
|
110 |
+
if self.official_model is not None:
|
111 |
+
try:
|
112 |
+
print("Falling back to official Whisper implementation...")
|
113 |
+
result = self.official_model.transcribe(audio_path)
|
114 |
+
transcription = result["text"]
|
115 |
+
print("Transcription successful with official implementation")
|
116 |
+
return transcription
|
117 |
+
except Exception as e:
|
118 |
+
print(f"Error in official Whisper transcription: {e}")
|
119 |
+
traceback.print_exc()
|
120 |
+
|
121 |
+
print("All transcription attempts failed")
|
122 |
+
return "Error: Transcription failed. Please check the logs for details."
|
123 |
+
|
124 |
+
class GrammarCorrector:
|
125 |
+
def __init__(self):
|
126 |
+
print("Initializing grammar corrector...")
|
127 |
+
try:
|
128 |
+
# Initialize grammar correction pipeline
|
129 |
+
self.corrector = pipeline("text2text-generation", model="pszemraj/flan-t5-large-grammar-synthesis")
|
130 |
+
print("Grammar corrector initialized successfully")
|
131 |
+
except Exception as e:
|
132 |
+
print(f"Error initializing grammar corrector: {e}")
|
133 |
+
traceback.print_exc()
|
134 |
+
self.corrector = None
|
135 |
+
|
136 |
+
def correct(self, text):
|
137 |
+
if not text or not text.strip():
|
138 |
+
return text
|
139 |
+
|
140 |
+
if self.corrector is not None:
|
141 |
+
try:
|
142 |
+
# Use the grammar correction pipeline
|
143 |
+
corrected_text = self.corrector(f"grammar correction: {text}")[0]['generated_text']
|
144 |
+
return corrected_text
|
145 |
+
except Exception as e:
|
146 |
+
print(f"Error in grammar correction: {e}")
|
147 |
+
return text
|
148 |
+
else:
|
149 |
+
print("No valid grammar correction model available. Returning original text.")
|
150 |
+
return text
|
151 |
+
|
152 |
+
class TextToSpeech:
|
153 |
+
def __init__(self):
|
154 |
+
print("Initializing text-to-speech engine...")
|
155 |
+
|
156 |
+
def speak(self, text, output_file="output_speech.mp3"):
|
157 |
+
try:
|
158 |
+
tts = gTTS(text=text, lang='en', slow=False)
|
159 |
+
tts.save(output_file)
|
160 |
+
print(f"Speech saved to {output_file}")
|
161 |
+
return output_file
|
162 |
+
except Exception as e:
|
163 |
+
print(f"Error with gTTS: {e}")
|
164 |
+
traceback.print_exc()
|
165 |
+
return False
|
166 |
+
|
167 |
+
class SpeechProcessor:
|
168 |
+
def __init__(self, whisper_model_size="tiny"):
|
169 |
+
print(f"Initializing Speech Processor with Whisper model size: {whisper_model_size}")
|
170 |
+
self.transcriber = WhisperTranscriber(model_size=whisper_model_size)
|
171 |
+
self.grammar_corrector = GrammarCorrector()
|
172 |
+
self.tts = TextToSpeech()
|
173 |
+
|
174 |
+
def process_text(self, text):
|
175 |
+
"""Process text input: correct grammar and generate speech"""
|
176 |
+
print("Processing text input...")
|
177 |
+
|
178 |
+
# Correct grammar and punctuation
|
179 |
+
corrected_text = self.grammar_corrector.correct(text)
|
180 |
+
|
181 |
+
# Generate speech from corrected text
|
182 |
+
speech_file = self.tts.speak(corrected_text, "output_speech.mp3")
|
183 |
+
|
184 |
+
return corrected_text, speech_file
|
185 |
+
|
186 |
+
def process_audio(self, audio_path):
|
187 |
+
"""Process audio input: transcribe, correct grammar, and generate speech"""
|
188 |
+
print(f"Processing audio input from: {audio_path}")
|
189 |
+
|
190 |
+
if not audio_path:
|
191 |
+
return "Failed to get audio", None, None
|
192 |
+
|
193 |
+
# Transcribe audio
|
194 |
+
transcription = self.transcriber.transcribe(audio_path)
|
195 |
+
|
196 |
+
if transcription.startswith("Error:"):
|
197 |
+
return transcription, None, None
|
198 |
+
|
199 |
+
# Correct grammar and punctuation
|
200 |
+
corrected_text = self.grammar_corrector.correct(transcription)
|
201 |
+
|
202 |
+
# Generate speech from corrected text
|
203 |
+
speech_file = self.tts.speak(corrected_text, "output_speech.mp3")
|
204 |
+
|
205 |
+
return transcription, corrected_text, speech_file
|
206 |
+
|
207 |
+
# Initialize the processor
|
208 |
+
processor = SpeechProcessor(whisper_model_size="tiny")
|
209 |
+
|
210 |
+
# Define Gradio functions for the interface
|
211 |
+
def process_text_input(text):
|
212 |
+
"""Handle text input from Gradio interface"""
|
213 |
+
corrected_text, speech_file = processor.process_text(text)
|
214 |
+
return corrected_text, speech_file
|
215 |
+
|
216 |
+
def process_audio_input(audio_file):
|
217 |
+
"""Handle audio upload/recording from Gradio interface"""
|
218 |
+
if audio_file is None:
|
219 |
+
return "No audio provided", "No audio provided", None
|
220 |
+
|
221 |
+
transcription, corrected_text, speech_file = processor.process_audio(audio_file)
|
222 |
+
|
223 |
+
if transcription.startswith("Error:"):
|
224 |
+
return transcription, "", None
|
225 |
+
|
226 |
+
return transcription, corrected_text, speech_file
|
227 |
+
|
228 |
+
# Create the Gradio interface
|
229 |
+
def create_gradio_interface():
|
230 |
+
with gr.Blocks(title="Speech Processing System") as demo:
|
231 |
+
gr.Markdown("# Speech Processing System")
|
232 |
+
gr.Markdown("Transcribe, correct grammar, and generate speech.")
|
233 |
+
|
234 |
+
with gr.Tab("Text Input"):
|
235 |
+
with gr.Row():
|
236 |
+
text_input = gr.Textbox(placeholder="Enter text to process", label="Input Text", lines=5)
|
237 |
+
|
238 |
+
text_button = gr.Button("Process Text")
|
239 |
+
|
240 |
+
with gr.Row():
|
241 |
+
corrected_text_output = gr.Textbox(label="Corrected Text", lines=5)
|
242 |
+
speech_output = gr.Audio(label="Speech Output")
|
243 |
+
|
244 |
+
text_button.click(
|
245 |
+
fn=process_text_input,
|
246 |
+
inputs=[text_input],
|
247 |
+
outputs=[corrected_text_output, speech_output]
|
248 |
+
)
|
249 |
+
|
250 |
+
with gr.Tab("Audio Input"):
|
251 |
+
with gr.Row():
|
252 |
+
audio_input = gr.Audio(
|
253 |
+
sources=["microphone", "upload"],
|
254 |
+
type="filepath",
|
255 |
+
label="Upload or Record Audio"
|
256 |
+
)
|
257 |
+
|
258 |
+
audio_button = gr.Button("Process Audio")
|
259 |
+
|
260 |
+
with gr.Row():
|
261 |
+
transcription_output = gr.Textbox(label="Transcription", lines=3)
|
262 |
+
audio_corrected_text = gr.Textbox(label="Corrected Text", lines=3)
|
263 |
+
|
264 |
+
with gr.Row():
|
265 |
+
audio_speech_output = gr.Audio(label="Speech Output")
|
266 |
+
|
267 |
+
audio_button.click(
|
268 |
+
fn=process_audio_input,
|
269 |
+
inputs=[audio_input],
|
270 |
+
outputs=[transcription_output, audio_corrected_text, audio_speech_output]
|
271 |
+
)
|
272 |
+
|
273 |
+
gr.Markdown("## How to use")
|
274 |
+
gr.Markdown("""
|
275 |
+
1. **Text Input Tab**: Enter text, click 'Process Text'. The system will correct grammar and generate speech.
|
276 |
+
2. **Audio Input Tab**: Upload an audio file or record using your microphone, then click 'Process Audio'.
|
277 |
+
The system will transcribe your speech, correct grammar, and generate improved speech.
|
278 |
+
""")
|
279 |
+
|
280 |
+
return demo
|
281 |
+
|
282 |
+
# Launch the interface
|
283 |
+
demo = create_gradio_interface()
|
284 |
+
|
285 |
+
if __name__ == "__main__":
|
286 |
+
demo.launch()
|