File size: 8,302 Bytes
8c6664d
 
 
 
c1e3042
8c6664d
c1e3042
 
 
8c6664d
c1e3042
 
 
 
8c6664d
 
 
c1e3042
8c6664d
c1e3042
 
8c6664d
 
 
 
 
 
 
 
 
 
c1e3042
8c6664d
c1e3042
8c6664d
 
 
 
 
 
 
c1e3042
 
 
 
 
 
 
99c2dd3
c1e3042
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8c6664d
 
 
 
 
 
 
 
 
 
 
 
 
c1e3042
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8c6664d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c1e3042
 
 
8c6664d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c1e3042
9d49df0
8c6664d
 
 
c1e3042
8c6664d
 
9d49df0
 
 
d059e20
9d49df0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
import gradio as gr
import torch
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
import numpy as np
import os

os.environ["OMP_NUM_THREADS"] = "2"
os.environ["MKL_NUM_THREADS"] = "2"
torch.set_num_threads(2)

device = "cpu"
torch_dtype = torch.float32

model_id = "openai/whisper-tiny"

model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id, 
    dtype=torch_dtype, 
    low_cpu_mem_usage=True, 
    use_safetensors=True,
    attn_implementation="sdpa"
)
model.to(device)

processor = AutoProcessor.from_pretrained(model_id)

pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    dtype=torch_dtype,
    device=device,
    ignore_warning=True
)

def transcribe_audio(audio_file, task="transcribe", language="auto", return_timestamps=False):
    if audio_file is None:
        return "No audio file provided."
    
    try:
        with torch.inference_mode():
            generate_kwargs = {
                "task": task,
                "language": None if language == "auto" else language,
                "num_beams": 1,
                "do_sample": False,
                "temperature": 0.0,
                "max_new_tokens": 220,
                "compression_ratio_threshold": 1.35,
                "logprob_threshold": -1.0,
                "no_speech_threshold": 0.6,
            }
            
            if task == "translate":
                generate_kwargs["task"] = "translate"
            
            result = pipe(
                audio_file, 
                return_timestamps=return_timestamps,
                generate_kwargs=generate_kwargs
            )
            
            if return_timestamps and "chunks" in result:
                formatted_result = []
                for chunk in result["chunks"]:
                    timestamp = f"[{chunk['timestamp'][0]:.2f}s - {chunk['timestamp'][1]:.2f}s]"
                    formatted_result.append(f"{timestamp} {chunk['text']}")
                return "\n".join(formatted_result)
            else:
                return result["text"]
            
    except Exception as e:
        return f"Error processing audio: {str(e)}"

def transcribe_microphone(audio_data, task="transcribe", language="auto", return_timestamps=False):
    if audio_data is None:
        return "No audio recorded."
    
    try:
        sample_rate, audio_array = audio_data
        audio_array = audio_array.astype(np.float32)
        audio_array = audio_array / np.max(np.abs(audio_array))
        
        with torch.inference_mode():
            generate_kwargs = {
                "task": task,
                "language": None if language == "auto" else language,
                "num_beams": 1,
                "do_sample": False,
                "temperature": 0.0,
                "max_new_tokens": 448,
                "compression_ratio_threshold": 1.35,
                "logprob_threshold": -1.0,
                "no_speech_threshold": 0.6,
            }
            
            if task == "translate":
                generate_kwargs["task"] = "translate"
            
            result = pipe(
                {"array": audio_array, "sampling_rate": sample_rate},
                return_timestamps=return_timestamps,
                generate_kwargs=generate_kwargs
            )
            
            if return_timestamps and "chunks" in result:
                formatted_result = []
                for chunk in result["chunks"]:
                    timestamp = f"[{chunk['timestamp'][0]:.2f}s - {chunk['timestamp'][1]:.2f}s]"
                    formatted_result.append(f"{timestamp} {chunk['text']}")
                return "\n".join(formatted_result)
            else:
                return result["text"]
            
    except Exception as e:
        return f"Error processing audio: {str(e)}"

languages = [
    ("Auto Detect", "auto"),
    ("English", "en"),
    ("Chinese", "zh"),
    ("German", "de"),
    ("Spanish", "es"),
    ("Russian", "ru"),
    ("Korean", "ko"),
    ("French", "fr"),
    ("Japanese", "ja"),
    ("Portuguese", "pt"),
    ("Turkish", "tr"),
    ("Polish", "pl"),
    ("Catalan", "ca"),
    ("Dutch", "nl"),
    ("Arabic", "ar"),
    ("Swedish", "sv"),
    ("Italian", "it"),
    ("Indonesian", "id"),
    ("Hindi", "hi"),
    ("Finnish", "fi"),
    ("Vietnamese", "vi"),
    ("Hebrew", "he"),
    ("Ukrainian", "uk"),
    ("Greek", "el"),
    ("Malay", "ms"),
    ("Czech", "cs"),
    ("Romanian", "ro"),
    ("Danish", "da"),
    ("Hungarian", "hu"),
    ("Tamil", "ta"),
    ("Norwegian", "no"),
    ("Thai", "th"),
    ("Urdu", "ur"),
    ("Croatian", "hr"),
    ("Bulgarian", "bg"),
    ("Lithuanian", "lt"),
    ("Latin", "la"),
]

with gr.Blocks(title="Whisper Tiny - Speech to Text") as demo:
    gr.Markdown("# 🎤 Whisper Tiny - Speech to Text")
    gr.Markdown("Upload an audio file or record directly to get fast transcription using OpenAI's Whisper Tiny model (39M parameters).")
    
    with gr.Tab("Upload Audio File"):
        with gr.Row():
            with gr.Column():
                audio_file = gr.Audio(
                    label="Upload Audio File",
                    type="filepath"
                )
                
                task_file = gr.Radio(
                    choices=[("Transcribe", "transcribe"), ("Translate to English", "translate")],
                    value="transcribe",
                    label="Task"
                )
                
                language_file = gr.Dropdown(
                    choices=languages,
                    value="auto",
                    label="Source Language"
                )
                
                timestamps_file = gr.Checkbox(
                    label="Return Timestamps",
                    value=False
                )
                
                submit_file = gr.Button("Transcribe Audio File", variant="primary")
            
            with gr.Column():
                output_file = gr.Textbox(
                    label="Transcription Result",
                    lines=10,
                    max_lines=20
                )
    
    with gr.Tab("Record Audio"):
        with gr.Row():
            with gr.Column():
                audio_mic = gr.Audio(
                    label="Record Audio",
                    sources=["microphone"]
                )
                
                task_mic = gr.Radio(
                    choices=[("Transcribe", "transcribe"), ("Translate to English", "translate")],
                    value="transcribe",
                    label="Task"
                )
                
                language_mic = gr.Dropdown(
                    choices=languages,
                    value="auto",
                    label="Source Language"
                )
                
                timestamps_mic = gr.Checkbox(
                    label="Return Timestamps",
                    value=False
                )
                
                submit_mic = gr.Button("Transcribe Recording", variant="primary")
            
            with gr.Column():
                output_mic = gr.Textbox(
                    label="Transcription Result",
                    lines=10,
                    max_lines=20
                )
    
    submit_file.click(
        transcribe_audio,
        inputs=[audio_file, task_file, language_file, timestamps_file],
        outputs=output_file
    )
    
    submit_mic.click(
        transcribe_microphone,
        inputs=[audio_mic, task_mic, language_mic, timestamps_mic],
        outputs=output_mic
    )
    
    gr.Markdown("### Features:")
    gr.Markdown("- **Lightweight**: Powered by Whisper Tiny model (39M parameters)")
    gr.Markdown("- **CPU Optimized**: Optimized for 2-core CPU with 16GB RAM")  
    gr.Markdown("- **Multi-language**: Supports 99+ languages")
    gr.Markdown("- **Translation**: Can translate speech to English")
    gr.Markdown("- **Timestamps**: Optional word-level or sentence-level timestamps")
    gr.Markdown("- **Fast Processing**: Smallest Whisper model for maximum speed")

if __name__ == "__main__":
    demo.launch(
        server_name="0.0.0.0", 
        server_port=7860,
        show_error=True
    )