File size: 8,717 Bytes
305c59b
7fe102a
16649a3
7fe102a
 
16649a3
ba6b40b
16649a3
7fe102a
 
 
 
 
 
ba6b40b
 
7fe102a
1e0f1bc
16649a3
 
 
 
 
 
 
 
 
7fe102a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cfd9ff1
 
 
1ba1d48
 
 
6b2690e
 
 
 
 
 
 
 
 
 
 
 
 
cfd9ff1
 
 
 
6b2690e
 
 
 
 
 
7fe102a
868debc
16649a3
7fe102a
 
 
 
 
16649a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7fe102a
16649a3
1ba1d48
ba6b40b
1ba1d48
 
 
 
 
 
 
 
ba6b40b
 
cfd9ff1
16649a3
ba6b40b
 
 
 
 
 
 
 
 
 
 
 
 
 
7fe102a
 
 
 
 
16649a3
1e0f1bc
7fe102a
868debc
6b2690e
 
 
 
 
 
 
868debc
 
 
 
 
7fe102a
 
 
868debc
 
7fe102a
 
16649a3
 
7fe102a
868debc
 
 
6b2690e
 
868debc
1e0f1bc
7fe102a
868debc
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
import gradio as gr
import whisper
import torch
import os
from pydub import AudioSegment
from transformers import pipeline
from faster_whisper import WhisperModel  # Import faster-whisper

# Mapping of model names to Whisper model sizes
MODELS = {
    "Tiny (Fastest)": "tiny",
    "Base (Faster)": "base",
    "Small (Balanced)": "small",
    "Medium (Accurate)": "medium",
    "Large (Most Accurate)": "large",
    "Systran Faster Whisper Large v3": "Systran/faster-whisper-large-v3"  # Add the new model
}

# Fine-tuned models for specific languages
FINE_TUNED_MODELS = {
    "Tamil": {
        "model": "vasista22/whisper-tamil-medium",
        "language": "ta"
    },
    # Add more fine-tuned models for other languages here
}

# Mapping of full language names to language codes
LANGUAGE_NAME_TO_CODE = {
    "Auto Detect": "Auto Detect",
    "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",
    "Maori": "mi",
    "Malayalam": "ml",
    "Welsh": "cy",
    "Slovak": "sk",
    "Telugu": "te",
    "Persian": "fa",
    "Latvian": "lv",
    "Bengali": "bn",
    "Serbian": "sr",
    "Azerbaijani": "az",
    "Slovenian": "sl",
    "Kannada": "kn",
    "Estonian": "et",
    "Macedonian": "mk",
    "Breton": "br",
    "Basque": "eu",
    "Icelandic": "is",
    "Armenian": "hy",
    "Nepali": "ne",
    "Mongolian": "mn",
    "Bosnian": "bs",
    "Kazakh": "kk",
    "Albanian": "sq",
    "Swahili": "sw",
    "Galician": "gl",
    "Marathi": "mr",
    "Punjabi": "pa",
    "Sinhala": "si",  # Sinhala support
    "Khmer": "km",
    "Shona": "sn",
    "Yoruba": "yo",
    "Somali": "so",
    "Afrikaans": "af",
    "Occitan": "oc",
    "Georgian": "ka",
    "Belarusian": "be",
    "Tajik": "tg",
    "Sindhi": "sd",
    "Gujarati": "gu",
    "Amharic": "am",
    "Yiddish": "yi",
    "Lao": "lo",
    "Uzbek": "uz",
    "Faroese": "fo",
    "Haitian Creole": "ht",
    "Pashto": "ps",
    "Turkmen": "tk",
    "Nynorsk": "nn",
    "Maltese": "mt",
    "Sanskrit": "sa",
    "Luxembourgish": "lb",
    "Burmese": "my",
    "Tibetan": "bo",
    "Tagalog": "tl",
    "Malagasy": "mg",
    "Assamese": "as",
    "Tatar": "tt",
    "Hawaiian": "haw",
    "Lingala": "ln",
    "Hausa": "ha",
    "Bashkir": "ba",
    "Javanese": "jw",
    "Sundanese": "su",
}

# Reverse mapping of language codes to full language names
CODE_TO_LANGUAGE_NAME = {v: k for k, v in LANGUAGE_NAME_TO_CODE.items()}

# Device and compute type for faster-whisper
device, torch_dtype = ("cuda", "float32") if torch.cuda.is_available() else ("cpu", "int8")

def detect_language(audio_file):
    """Detect the language of the audio file."""
    # Load the Whisper model (use "base" for faster detection)
    model = whisper.load_model("base")
    
    # Convert audio to 16kHz mono for better compatibility with Whisper
    audio = AudioSegment.from_file(audio_file)
    audio = audio.set_frame_rate(16000).set_channels(1)
    processed_audio_path = "processed_audio.wav"
    audio.export(processed_audio_path, format="wav")
    
    # Detect the language
    result = model.transcribe(processed_audio_path, task="detect_language", fp16=False)
    detected_language_code = result.get("language", "unknown")
    
    # Get the full language name from the code
    detected_language = CODE_TO_LANGUAGE_NAME.get(detected_language_code, "Unknown Language")
    
    # Clean up processed audio file
    os.remove(processed_audio_path)
    
    return f"Detected Language: {detected_language}"

def transcribe_audio(audio_file, language="Auto Detect", model_size="Base (Faster)"):
    """Transcribe the audio file."""
    # Convert audio to 16kHz mono for better compatibility
    audio = AudioSegment.from_file(audio_file)
    audio = audio.set_frame_rate(16000).set_channels(1)
    processed_audio_path = "processed_audio.wav"
    audio.export(processed_audio_path, format="wav")
    
    # Load the appropriate model
    if language in FINE_TUNED_MODELS:
        # Use the fine-tuned Whisper model for the selected language
        device = "cuda:0" if torch.cuda.is_available() else "cpu"
        transcribe = pipeline(
            task="automatic-speech-recognition",
            model=FINE_TUNED_MODELS[language]["model"],
            chunk_length_s=30,
            device=device
        )
        transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(
            language=FINE_TUNED_MODELS[language]["language"],
            task="transcribe"
        )
        result = transcribe(processed_audio_path)
        transcription = result["text"]
        detected_language = language
    else:
        # Use the selected Whisper model
        if model_size == "Systran Faster Whisper Large v3":
            # Use faster-whisper for the Systran model
            model = WhisperModel(MODELS[model_size], device=device, compute_type=torch_dtype)
            segments, info = model.transcribe(
                processed_audio_path,
                task="transcribe",
                word_timestamps=True,
                repetition_penalty=1.1,
                temperature=[0.0, 0.1, 0.2, 0.3, 0.4, 0.6, 0.8, 1.0],
            )
            transcription = " ".join([segment.text for segment in segments])
            detected_language_code = info.language
            detected_language = CODE_TO_LANGUAGE_NAME.get(detected_language_code, "Unknown Language")
        else:
            # Use the standard Whisper model
            model = whisper.load_model(MODELS[model_size])
            
            # Transcribe the audio
            if language == "Auto Detect":
                result = model.transcribe(processed_audio_path, fp16=False)  # Auto-detect language
                detected_language_code = result.get("language", "unknown")
                detected_language = CODE_TO_LANGUAGE_NAME.get(detected_language_code, "Unknown Language")
            else:
                language_code = LANGUAGE_NAME_TO_CODE.get(language, "en")  # Default to English if not found
                result = model.transcribe(processed_audio_path, language=language_code, fp16=False)
                detected_language = language
            
            transcription = result["text"]
    
    # Clean up processed audio file
    os.remove(processed_audio_path)
    
    # Return transcription and detected language
    return f"Detected Language: {detected_language}\n\nTranscription:\n{transcription}"

# Define the Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# Audio Transcription and Language Detection")
    
    with gr.Tab("Detect Language"):
        gr.Markdown("Upload an audio file to detect its language.")
        detect_audio_input = gr.Audio(type="filepath", label="Upload Audio File")
        detect_language_output = gr.Textbox(label="Detected Language")
        detect_button = gr.Button("Detect Language")
    
    with gr.Tab("Transcribe Audio"):
        gr.Markdown("Upload an audio file, select a language (or choose 'Auto Detect'), and choose a model for transcription.")
        transcribe_audio_input = gr.Audio(type="filepath", label="Upload Audio File")
        language_dropdown = gr.Dropdown(
            choices=list(LANGUAGE_NAME_TO_CODE.keys()),  # Full language names
            label="Select Language",
            value="Auto Detect"
        )
        model_dropdown = gr.Dropdown(
            choices=list(MODELS.keys()),  # Model options
            label="Select Model",
            value="Base (Faster)",  # Default to "Base" model
            interactive=True  # Allow model selection by default
        )
        transcribe_output = gr.Textbox(label="Transcription and Detected Language")
        transcribe_button = gr.Button("Transcribe Audio")
    
    # Link buttons to functions
    detect_button.click(detect_language, inputs=detect_audio_input, outputs=detect_language_output)
    transcribe_button.click(transcribe_audio, inputs=[transcribe_audio_input, language_dropdown, model_dropdown], outputs=transcribe_output)

# Launch the Gradio interface
demo.launch()