Update app.py
Browse files
app.py
CHANGED
@@ -1,694 +1,694 @@
|
|
1 |
-
# Copyright (c) Meta Platforms, Inc. and affiliates
|
2 |
-
# All rights reserved.
|
3 |
-
#
|
4 |
-
# This source code is licensed under the license found in the
|
5 |
-
# LICENSE file in the root directory of this source tree.
|
6 |
-
|
7 |
-
from __future__ import annotations
|
8 |
-
|
9 |
-
import gradio as gr
|
10 |
-
import numpy as np
|
11 |
-
# import torch
|
12 |
-
|
13 |
-
|
14 |
-
from gradio_client import Client
|
15 |
-
|
16 |
-
client = Client("https://facebook-seamless-m4t.hf.space/")
|
17 |
-
|
18 |
-
DESCRIPTION = """
|
19 |
-
|
20 |
-
# SM4T
|
21 |
-
|
22 |
-
Ứng dụng có thể chuyển đổi giọng nói hoặc chữ viết sang giọng nói hoặc chữ viết của một ngôn ngữ khác.
|
23 |
-
\nHiện tại SM4T đã hỗ trợ 94 ngôn ngữ khác nhau.
|
24 |
-
|
25 |
-
"""
|
26 |
-
|
27 |
-
TASK_NAMES = [
|
28 |
-
"S2ST (Speech to Speech translation)",
|
29 |
-
"S2TT (Speech to Text translation)",
|
30 |
-
"T2ST (Text to Speech translation)",
|
31 |
-
"T2TT (Text to Text translation)",
|
32 |
-
"ASR (Automatic Speech Recognition)",
|
33 |
-
]
|
34 |
-
|
35 |
-
# Language dict
|
36 |
-
language_code_to_name = {
|
37 |
-
"afr": "Afrikaans",
|
38 |
-
"amh": "Amharic",
|
39 |
-
"arb": "Modern Standard Arabic",
|
40 |
-
"ary": "Moroccan Arabic",
|
41 |
-
"arz": "Egyptian Arabic",
|
42 |
-
"asm": "Assamese",
|
43 |
-
"ast": "Asturian",
|
44 |
-
"azj": "North Azerbaijani",
|
45 |
-
"bel": "Belarusian",
|
46 |
-
"ben": "Bengali",
|
47 |
-
"bos": "Bosnian",
|
48 |
-
"bul": "Bulgarian",
|
49 |
-
"cat": "Catalan",
|
50 |
-
"ceb": "Cebuano",
|
51 |
-
"ces": "Czech",
|
52 |
-
"ckb": "Central Kurdish",
|
53 |
-
"cmn": "Mandarin Chinese",
|
54 |
-
"cym": "Welsh",
|
55 |
-
"dan": "Danish",
|
56 |
-
"deu": "German",
|
57 |
-
"ell": "Greek",
|
58 |
-
"eng": "English",
|
59 |
-
"est": "Estonian",
|
60 |
-
"eus": "Basque",
|
61 |
-
"fin": "Finnish",
|
62 |
-
"fra": "French",
|
63 |
-
"gaz": "West Central Oromo",
|
64 |
-
"gle": "Irish",
|
65 |
-
"glg": "Galician",
|
66 |
-
"guj": "Gujarati",
|
67 |
-
"heb": "Hebrew",
|
68 |
-
"hin": "Hindi",
|
69 |
-
"hrv": "Croatian",
|
70 |
-
"hun": "Hungarian",
|
71 |
-
"hye": "Armenian",
|
72 |
-
"ibo": "Igbo",
|
73 |
-
"ind": "Indonesian",
|
74 |
-
"isl": "Icelandic",
|
75 |
-
"ita": "Italian",
|
76 |
-
"jav": "Javanese",
|
77 |
-
"jpn": "Japanese",
|
78 |
-
"kam": "Kamba",
|
79 |
-
"kan": "Kannada",
|
80 |
-
"kat": "Georgian",
|
81 |
-
"kaz": "Kazakh",
|
82 |
-
"kea": "Kabuverdianu",
|
83 |
-
"khk": "Halh Mongolian",
|
84 |
-
"khm": "Khmer",
|
85 |
-
"kir": "Kyrgyz",
|
86 |
-
"kor": "Korean",
|
87 |
-
"lao": "Lao",
|
88 |
-
"lit": "Lithuanian",
|
89 |
-
"ltz": "Luxembourgish",
|
90 |
-
"lug": "Ganda",
|
91 |
-
"luo": "Luo",
|
92 |
-
"lvs": "Standard Latvian",
|
93 |
-
"mai": "Maithili",
|
94 |
-
"mal": "Malayalam",
|
95 |
-
"mar": "Marathi",
|
96 |
-
"mkd": "Macedonian",
|
97 |
-
"mlt": "Maltese",
|
98 |
-
"mni": "Meitei",
|
99 |
-
"mya": "Burmese",
|
100 |
-
"nld": "Dutch",
|
101 |
-
"nno": "Norwegian Nynorsk",
|
102 |
-
"nob": "Norwegian Bokm\u00e5l",
|
103 |
-
"npi": "Nepali",
|
104 |
-
"nya": "Nyanja",
|
105 |
-
"oci": "Occitan",
|
106 |
-
"ory": "Odia",
|
107 |
-
"pan": "Punjabi",
|
108 |
-
"pbt": "Southern Pashto",
|
109 |
-
"pes": "Western Persian",
|
110 |
-
"pol": "Polish",
|
111 |
-
"por": "Portuguese",
|
112 |
-
"ron": "Romanian",
|
113 |
-
"rus": "Russian",
|
114 |
-
"slk": "Slovak",
|
115 |
-
"slv": "Slovenian",
|
116 |
-
"sna": "Shona",
|
117 |
-
"snd": "Sindhi",
|
118 |
-
"som": "Somali",
|
119 |
-
"spa": "Spanish",
|
120 |
-
"srp": "Serbian",
|
121 |
-
"swe": "Swedish",
|
122 |
-
"swh": "Swahili",
|
123 |
-
"tam": "Tamil",
|
124 |
-
"tel": "Telugu",
|
125 |
-
"tgk": "Tajik",
|
126 |
-
"tgl": "Tagalog",
|
127 |
-
"tha": "Thai",
|
128 |
-
"tur": "Turkish",
|
129 |
-
"ukr": "Ukrainian",
|
130 |
-
"urd": "Urdu",
|
131 |
-
"uzn": "Northern Uzbek",
|
132 |
-
"vie": "Vietnamese",
|
133 |
-
"xho": "Xhosa",
|
134 |
-
"yor": "Yoruba",
|
135 |
-
"yue": "Cantonese",
|
136 |
-
"zlm": "Colloquial Malay",
|
137 |
-
"zsm": "Standard Malay",
|
138 |
-
"zul": "Zulu",
|
139 |
-
}
|
140 |
-
LANGUAGE_NAME_TO_CODE = {v: k for k, v in language_code_to_name.items()}
|
141 |
-
|
142 |
-
# Source langs: S2ST / S2TT / ASR don't need source lang
|
143 |
-
# T2TT / T2ST use this
|
144 |
-
text_source_language_codes = [
|
145 |
-
"afr",
|
146 |
-
"amh",
|
147 |
-
"arb",
|
148 |
-
"ary",
|
149 |
-
"arz",
|
150 |
-
"asm",
|
151 |
-
"azj",
|
152 |
-
"bel",
|
153 |
-
"ben",
|
154 |
-
"bos",
|
155 |
-
"bul",
|
156 |
-
"cat",
|
157 |
-
"ceb",
|
158 |
-
"ces",
|
159 |
-
"ckb",
|
160 |
-
"cmn",
|
161 |
-
"cym",
|
162 |
-
"dan",
|
163 |
-
"deu",
|
164 |
-
"ell",
|
165 |
-
"eng",
|
166 |
-
"est",
|
167 |
-
"eus",
|
168 |
-
"fin",
|
169 |
-
"fra",
|
170 |
-
"gaz",
|
171 |
-
"gle",
|
172 |
-
"glg",
|
173 |
-
"guj",
|
174 |
-
"heb",
|
175 |
-
"hin",
|
176 |
-
"hrv",
|
177 |
-
"hun",
|
178 |
-
"hye",
|
179 |
-
"ibo",
|
180 |
-
"ind",
|
181 |
-
"isl",
|
182 |
-
"ita",
|
183 |
-
"jav",
|
184 |
-
"jpn",
|
185 |
-
"kan",
|
186 |
-
"kat",
|
187 |
-
"kaz",
|
188 |
-
"khk",
|
189 |
-
"khm",
|
190 |
-
"kir",
|
191 |
-
"kor",
|
192 |
-
"lao",
|
193 |
-
"lit",
|
194 |
-
"lug",
|
195 |
-
"luo",
|
196 |
-
"lvs",
|
197 |
-
"mai",
|
198 |
-
"mal",
|
199 |
-
"mar",
|
200 |
-
"mkd",
|
201 |
-
"mlt",
|
202 |
-
"mni",
|
203 |
-
"mya",
|
204 |
-
"nld",
|
205 |
-
"nno",
|
206 |
-
"nob",
|
207 |
-
"npi",
|
208 |
-
"nya",
|
209 |
-
"ory",
|
210 |
-
"pan",
|
211 |
-
"pbt",
|
212 |
-
"pes",
|
213 |
-
"pol",
|
214 |
-
"por",
|
215 |
-
"ron",
|
216 |
-
"rus",
|
217 |
-
"slk",
|
218 |
-
"slv",
|
219 |
-
"sna",
|
220 |
-
"snd",
|
221 |
-
"som",
|
222 |
-
"spa",
|
223 |
-
"srp",
|
224 |
-
"swe",
|
225 |
-
"swh",
|
226 |
-
"tam",
|
227 |
-
"tel",
|
228 |
-
"tgk",
|
229 |
-
"tgl",
|
230 |
-
"tha",
|
231 |
-
"tur",
|
232 |
-
"ukr",
|
233 |
-
"urd",
|
234 |
-
"uzn",
|
235 |
-
"vie",
|
236 |
-
"yor",
|
237 |
-
"yue",
|
238 |
-
"zsm",
|
239 |
-
"zul",
|
240 |
-
]
|
241 |
-
TEXT_SOURCE_LANGUAGE_NAMES = sorted(
|
242 |
-
[language_code_to_name[code] for code in text_source_language_codes]
|
243 |
-
)
|
244 |
-
|
245 |
-
# Target langs:
|
246 |
-
# S2ST / T2ST
|
247 |
-
s2st_target_language_codes = [
|
248 |
-
"eng",
|
249 |
-
"arb",
|
250 |
-
"ben",
|
251 |
-
"cat",
|
252 |
-
"ces",
|
253 |
-
"cmn",
|
254 |
-
"cym",
|
255 |
-
"dan",
|
256 |
-
"deu",
|
257 |
-
"est",
|
258 |
-
"fin",
|
259 |
-
"fra",
|
260 |
-
"hin",
|
261 |
-
"ind",
|
262 |
-
"ita",
|
263 |
-
"jpn",
|
264 |
-
"kor",
|
265 |
-
"mlt",
|
266 |
-
"nld",
|
267 |
-
"pes",
|
268 |
-
"pol",
|
269 |
-
"por",
|
270 |
-
"ron",
|
271 |
-
"rus",
|
272 |
-
"slk",
|
273 |
-
"spa",
|
274 |
-
"swe",
|
275 |
-
"swh",
|
276 |
-
"tel",
|
277 |
-
"tgl",
|
278 |
-
"tha",
|
279 |
-
"tur",
|
280 |
-
"ukr",
|
281 |
-
"urd",
|
282 |
-
"uzn",
|
283 |
-
"vie",
|
284 |
-
]
|
285 |
-
S2ST_TARGET_LANGUAGE_NAMES = sorted(
|
286 |
-
[language_code_to_name[code] for code in s2st_target_language_codes]
|
287 |
-
)
|
288 |
-
# S2TT / ASR
|
289 |
-
S2TT_TARGET_LANGUAGE_NAMES = TEXT_SOURCE_LANGUAGE_NAMES
|
290 |
-
# T2TT
|
291 |
-
T2TT_TARGET_LANGUAGE_NAMES = TEXT_SOURCE_LANGUAGE_NAMES
|
292 |
-
|
293 |
-
# Download sample input audio files
|
294 |
-
filenames = ["assets/sample_input.mp3", "assets/sample_input_2.mp3"]
|
295 |
-
# for filename in filenames:
|
296 |
-
# hf_hub_download(
|
297 |
-
# repo_id="facebook/seamless_m4t",
|
298 |
-
# repo_type="space",
|
299 |
-
# filename=filename,
|
300 |
-
# local_dir=".",
|
301 |
-
# )
|
302 |
-
|
303 |
-
AUDIO_SAMPLE_RATE = 16000.0
|
304 |
-
MAX_INPUT_AUDIO_LENGTH = 60 # in seconds
|
305 |
-
DEFAULT_TARGET_LANGUAGE = "French"
|
306 |
-
|
307 |
-
# device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
308 |
-
|
309 |
-
def api_predict(
|
310 |
-
task_name: str,
|
311 |
-
audio_source: str,
|
312 |
-
input_audio_mic: str | None,
|
313 |
-
input_audio_file: str | None,
|
314 |
-
input_text: str | None,
|
315 |
-
source_language: str | None,
|
316 |
-
target_language: str,):
|
317 |
-
|
318 |
-
audio_out, text_out = client.predict(task_name,
|
319 |
-
audio_source,
|
320 |
-
input_audio_mic,
|
321 |
-
input_audio_file,
|
322 |
-
input_text,
|
323 |
-
source_language,
|
324 |
-
target_language,
|
325 |
-
api_name="/run")
|
326 |
-
return audio_out, text_out
|
327 |
-
|
328 |
-
|
329 |
-
|
330 |
-
|
331 |
-
|
332 |
-
def process_s2st_example(
|
333 |
-
input_audio_file: str, target_language: str
|
334 |
-
) -> tuple[tuple[int, np.ndarray] | None, str]:
|
335 |
-
return api_predict(
|
336 |
-
task_name="S2ST",
|
337 |
-
audio_source="file",
|
338 |
-
input_audio_mic=None,
|
339 |
-
input_audio_file=input_audio_file,
|
340 |
-
input_text=None,
|
341 |
-
source_language=None,
|
342 |
-
target_language=target_language,
|
343 |
-
)
|
344 |
-
|
345 |
-
|
346 |
-
def process_s2tt_example(
|
347 |
-
input_audio_file: str, target_language: str
|
348 |
-
) -> tuple[tuple[int, np.ndarray] | None, str]:
|
349 |
-
return api_predict(
|
350 |
-
task_name="S2TT",
|
351 |
-
audio_source="file",
|
352 |
-
input_audio_mic=None,
|
353 |
-
input_audio_file=input_audio_file,
|
354 |
-
input_text=None,
|
355 |
-
source_language=None,
|
356 |
-
target_language=target_language,
|
357 |
-
)
|
358 |
-
|
359 |
-
|
360 |
-
def process_t2st_example(
|
361 |
-
input_text: str, source_language: str, target_language: str
|
362 |
-
) -> tuple[tuple[int, np.ndarray] | None, str]:
|
363 |
-
return api_predict(
|
364 |
-
task_name="T2ST",
|
365 |
-
audio_source="",
|
366 |
-
input_audio_mic=None,
|
367 |
-
input_audio_file=None,
|
368 |
-
input_text=input_text,
|
369 |
-
source_language=source_language,
|
370 |
-
target_language=target_language,
|
371 |
-
)
|
372 |
-
|
373 |
-
|
374 |
-
def process_t2tt_example(
|
375 |
-
input_text: str, source_language: str, target_language: str
|
376 |
-
) -> tuple[tuple[int, np.ndarray] | None, str]:
|
377 |
-
return api_predict(
|
378 |
-
task_name="T2TT",
|
379 |
-
audio_source="",
|
380 |
-
input_audio_mic=None,
|
381 |
-
input_audio_file=None,
|
382 |
-
input_text=input_text,
|
383 |
-
source_language=source_language,
|
384 |
-
target_language=target_language,
|
385 |
-
)
|
386 |
-
|
387 |
-
|
388 |
-
def process_asr_example(
|
389 |
-
input_audio_file: str, target_language: str
|
390 |
-
) -> tuple[tuple[int, np.ndarray] | None, str]:
|
391 |
-
return api_predict(
|
392 |
-
task_name="ASR",
|
393 |
-
audio_source="file",
|
394 |
-
input_audio_mic=None,
|
395 |
-
input_audio_file=input_audio_file,
|
396 |
-
input_text=None,
|
397 |
-
source_language=None,
|
398 |
-
target_language=target_language,
|
399 |
-
)
|
400 |
-
|
401 |
-
|
402 |
-
def update_audio_ui(audio_source: str) -> tuple[dict, dict]:
|
403 |
-
mic = audio_source == "microphone"
|
404 |
-
return (
|
405 |
-
gr.update(visible=mic, value=None), # input_audio_mic
|
406 |
-
gr.update(visible=not mic, value=None), # input_audio_file
|
407 |
-
)
|
408 |
-
|
409 |
-
|
410 |
-
def update_input_ui(task_name: str) -> tuple[dict, dict, dict, dict]:
|
411 |
-
task_name = task_name.split()[0]
|
412 |
-
if task_name == "S2ST":
|
413 |
-
return (
|
414 |
-
gr.update(visible=True), # audio_box
|
415 |
-
gr.update(visible=False), # input_text
|
416 |
-
gr.update(visible=False), # source_language
|
417 |
-
gr.update(
|
418 |
-
visible=True,
|
419 |
-
choices=S2ST_TARGET_LANGUAGE_NAMES,
|
420 |
-
value=DEFAULT_TARGET_LANGUAGE,
|
421 |
-
), # target_language
|
422 |
-
)
|
423 |
-
elif task_name == "S2TT":
|
424 |
-
return (
|
425 |
-
gr.update(visible=True), # audio_box
|
426 |
-
gr.update(visible=False), # input_text
|
427 |
-
gr.update(visible=False), # source_language
|
428 |
-
gr.update(
|
429 |
-
visible=True,
|
430 |
-
choices=S2TT_TARGET_LANGUAGE_NAMES,
|
431 |
-
value=DEFAULT_TARGET_LANGUAGE,
|
432 |
-
), # target_language
|
433 |
-
)
|
434 |
-
elif task_name == "T2ST":
|
435 |
-
return (
|
436 |
-
gr.update(visible=False), # audio_box
|
437 |
-
gr.update(visible=True), # input_text
|
438 |
-
gr.update(visible=True), # source_language
|
439 |
-
gr.update(
|
440 |
-
visible=True,
|
441 |
-
choices=S2ST_TARGET_LANGUAGE_NAMES,
|
442 |
-
value=DEFAULT_TARGET_LANGUAGE,
|
443 |
-
), # target_language
|
444 |
-
)
|
445 |
-
elif task_name == "T2TT":
|
446 |
-
return (
|
447 |
-
gr.update(visible=False), # audio_box
|
448 |
-
gr.update(visible=True), # input_text
|
449 |
-
gr.update(visible=True), # source_language
|
450 |
-
gr.update(
|
451 |
-
visible=True,
|
452 |
-
choices=T2TT_TARGET_LANGUAGE_NAMES,
|
453 |
-
value=DEFAULT_TARGET_LANGUAGE,
|
454 |
-
), # target_language
|
455 |
-
)
|
456 |
-
elif task_name == "ASR":
|
457 |
-
return (
|
458 |
-
gr.update(visible=True), # audio_box
|
459 |
-
gr.update(visible=False), # input_text
|
460 |
-
gr.update(visible=False), # source_language
|
461 |
-
gr.update(
|
462 |
-
visible=True,
|
463 |
-
choices=S2TT_TARGET_LANGUAGE_NAMES,
|
464 |
-
value=DEFAULT_TARGET_LANGUAGE,
|
465 |
-
), # target_language
|
466 |
-
)
|
467 |
-
else:
|
468 |
-
raise ValueError(f"Unknown task: {task_name}")
|
469 |
-
|
470 |
-
|
471 |
-
def update_output_ui(task_name: str) -> tuple[dict, dict]:
|
472 |
-
task_name = task_name.split()[0]
|
473 |
-
if task_name in ["S2ST", "T2ST"]:
|
474 |
-
return (
|
475 |
-
gr.update(visible=True, value=None), # output_audio
|
476 |
-
gr.update(value=None), # output_text
|
477 |
-
)
|
478 |
-
elif task_name in ["S2TT", "T2TT", "ASR"]:
|
479 |
-
return (
|
480 |
-
gr.update(visible=False, value=None), # output_audio
|
481 |
-
gr.update(value=None), # output_text
|
482 |
-
)
|
483 |
-
else:
|
484 |
-
raise ValueError(f"Unknown task: {task_name}")
|
485 |
-
|
486 |
-
|
487 |
-
def update_example_ui(task_name: str) -> tuple[dict, dict, dict, dict, dict]:
|
488 |
-
task_name = task_name.split()[0]
|
489 |
-
return (
|
490 |
-
gr.update(visible=task_name == "S2ST"), # s2st_example_row
|
491 |
-
gr.update(visible=task_name == "S2TT"), # s2tt_example_row
|
492 |
-
gr.update(visible=task_name == "T2ST"), # t2st_example_row
|
493 |
-
gr.update(visible=task_name == "T2TT"), # t2tt_example_row
|
494 |
-
gr.update(visible=task_name == "ASR"), # asr_example_row
|
495 |
-
)
|
496 |
-
|
497 |
-
|
498 |
-
css = """
|
499 |
-
h1 {
|
500 |
-
text-align: center;
|
501 |
-
}
|
502 |
-
|
503 |
-
|
504 |
-
max-width: 730px;
|
505 |
-
margin: auto;
|
506 |
-
padding-top: 1.5rem;
|
507 |
-
}
|
508 |
-
"""
|
509 |
-
|
510 |
-
with gr.Blocks(css=css) as demo:
|
511 |
-
gr.Markdown(DESCRIPTION)
|
512 |
-
with gr.Group():
|
513 |
-
task_name = gr.Dropdown(
|
514 |
-
label="Task",
|
515 |
-
choices=TASK_NAMES,
|
516 |
-
value=TASK_NAMES[0],
|
517 |
-
)
|
518 |
-
with gr.Row():
|
519 |
-
source_language = gr.Dropdown(
|
520 |
-
label="Source language",
|
521 |
-
choices=TEXT_SOURCE_LANGUAGE_NAMES,
|
522 |
-
value="English",
|
523 |
-
visible=False,
|
524 |
-
)
|
525 |
-
target_language = gr.Dropdown(
|
526 |
-
label="Target language",
|
527 |
-
choices=S2ST_TARGET_LANGUAGE_NAMES,
|
528 |
-
value=DEFAULT_TARGET_LANGUAGE,
|
529 |
-
)
|
530 |
-
with gr.Row() as audio_box:
|
531 |
-
audio_source = gr.Radio(
|
532 |
-
label="Audio source",
|
533 |
-
choices=["file", "microphone"],
|
534 |
-
value="file",
|
535 |
-
)
|
536 |
-
input_audio_mic = gr.Audio(
|
537 |
-
label="Input speech",
|
538 |
-
type="filepath",
|
539 |
-
source="microphone",
|
540 |
-
visible=False,
|
541 |
-
)
|
542 |
-
input_audio_file = gr.Audio(
|
543 |
-
label="Input speech",
|
544 |
-
type="filepath",
|
545 |
-
source="upload",
|
546 |
-
visible=True,
|
547 |
-
)
|
548 |
-
input_text = gr.Textbox(label="Input text", visible=False)
|
549 |
-
with gr.Row():
|
550 |
-
btn = gr.Button("Translate")
|
551 |
-
btn_clean = gr.ClearButton([input_audio_mic, input_audio_file])
|
552 |
-
# gr.Markdown("## Text Examples")
|
553 |
-
with gr.Column():
|
554 |
-
output_audio = gr.Audio(
|
555 |
-
label="Translated speech",
|
556 |
-
autoplay=False,
|
557 |
-
streaming=False,
|
558 |
-
type="numpy",
|
559 |
-
)
|
560 |
-
output_text = gr.Textbox(label="Translated text")
|
561 |
-
|
562 |
-
with gr.Row(visible=True) as s2st_example_row:
|
563 |
-
s2st_examples = gr.Examples(
|
564 |
-
examples=[
|
565 |
-
["assets/sample_input.mp3", "French"],
|
566 |
-
["assets/sample_input.mp3", "Mandarin Chinese"],
|
567 |
-
["assets/sample_input_2.mp3", "Hindi"],
|
568 |
-
["assets/sample_input_2.mp3", "Spanish"],
|
569 |
-
],
|
570 |
-
inputs=[input_audio_file, target_language],
|
571 |
-
outputs=[output_audio, output_text],
|
572 |
-
fn=process_s2st_example,
|
573 |
-
)
|
574 |
-
with gr.Row(visible=False) as s2tt_example_row:
|
575 |
-
s2tt_examples = gr.Examples(
|
576 |
-
examples=[
|
577 |
-
["assets/sample_input.mp3", "French"],
|
578 |
-
["assets/sample_input.mp3", "Mandarin Chinese"],
|
579 |
-
["assets/sample_input_2.mp3", "Hindi"],
|
580 |
-
["assets/sample_input_2.mp3", "Spanish"],
|
581 |
-
],
|
582 |
-
inputs=[input_audio_file, target_language],
|
583 |
-
outputs=[output_audio, output_text],
|
584 |
-
fn=process_s2tt_example,
|
585 |
-
)
|
586 |
-
with gr.Row(visible=False) as t2st_example_row:
|
587 |
-
t2st_examples = gr.Examples(
|
588 |
-
examples=[
|
589 |
-
["My favorite animal is the elephant.", "English", "French"],
|
590 |
-
["My favorite animal is the elephant.", "English", "Mandarin Chinese"],
|
591 |
-
[
|
592 |
-
"Meta AI's Seamless M4T model is democratising spoken communication across language barriers",
|
593 |
-
"English",
|
594 |
-
"Hindi",
|
595 |
-
],
|
596 |
-
[
|
597 |
-
"Meta AI's Seamless M4T model is democratising spoken communication across language barriers",
|
598 |
-
"English",
|
599 |
-
"Spanish",
|
600 |
-
],
|
601 |
-
],
|
602 |
-
inputs=[input_text, source_language, target_language],
|
603 |
-
outputs=[output_audio, output_text],
|
604 |
-
fn=process_t2st_example,
|
605 |
-
)
|
606 |
-
with gr.Row(visible=False) as t2tt_example_row:
|
607 |
-
t2tt_examples = gr.Examples(
|
608 |
-
examples=[
|
609 |
-
["My favorite animal is the elephant.", "English", "French"],
|
610 |
-
["My favorite animal is the elephant.", "English", "Mandarin Chinese"],
|
611 |
-
[
|
612 |
-
"Meta AI's Seamless M4T model is democratising spoken communication across language barriers",
|
613 |
-
"English",
|
614 |
-
"Hindi",
|
615 |
-
],
|
616 |
-
[
|
617 |
-
"Meta AI's Seamless M4T model is democratising spoken communication across language barriers",
|
618 |
-
"English",
|
619 |
-
"Spanish",
|
620 |
-
],
|
621 |
-
],
|
622 |
-
inputs=[input_text, source_language, target_language],
|
623 |
-
outputs=[output_audio, output_text],
|
624 |
-
fn=process_t2tt_example,
|
625 |
-
)
|
626 |
-
with gr.Row(visible=False) as asr_example_row:
|
627 |
-
asr_examples = gr.Examples(
|
628 |
-
examples=[
|
629 |
-
["assets/sample_input.mp3", "English"],
|
630 |
-
["assets/sample_input_2.mp3", "English"],
|
631 |
-
],
|
632 |
-
inputs=[input_audio_file, target_language],
|
633 |
-
outputs=[output_audio, output_text],
|
634 |
-
fn=process_asr_example,
|
635 |
-
)
|
636 |
-
|
637 |
-
audio_source.change(
|
638 |
-
fn=update_audio_ui,
|
639 |
-
inputs=audio_source,
|
640 |
-
outputs=[
|
641 |
-
input_audio_mic,
|
642 |
-
input_audio_file,
|
643 |
-
],
|
644 |
-
queue=False,
|
645 |
-
api_name=False,
|
646 |
-
)
|
647 |
-
task_name.change(
|
648 |
-
fn=update_input_ui,
|
649 |
-
inputs=task_name,
|
650 |
-
outputs=[
|
651 |
-
audio_box,
|
652 |
-
input_text,
|
653 |
-
source_language,
|
654 |
-
target_language,
|
655 |
-
],
|
656 |
-
queue=False,
|
657 |
-
api_name=False,
|
658 |
-
).then(
|
659 |
-
fn=update_output_ui,
|
660 |
-
inputs=task_name,
|
661 |
-
outputs=[output_audio, output_text],
|
662 |
-
queue=False,
|
663 |
-
api_name=False,
|
664 |
-
).then(
|
665 |
-
fn=update_example_ui,
|
666 |
-
inputs=task_name,
|
667 |
-
outputs=[
|
668 |
-
s2st_example_row,
|
669 |
-
s2tt_example_row,
|
670 |
-
t2st_example_row,
|
671 |
-
t2tt_example_row,
|
672 |
-
asr_example_row,
|
673 |
-
],
|
674 |
-
queue=False,
|
675 |
-
api_name=False,
|
676 |
-
)
|
677 |
-
|
678 |
-
btn.click(
|
679 |
-
fn=api_predict,
|
680 |
-
inputs=[
|
681 |
-
task_name,
|
682 |
-
audio_source,
|
683 |
-
input_audio_mic,
|
684 |
-
input_audio_file,
|
685 |
-
input_text,
|
686 |
-
source_language,
|
687 |
-
target_language,
|
688 |
-
],
|
689 |
-
outputs=[output_audio, output_text],
|
690 |
-
api_name="run",
|
691 |
-
)
|
692 |
-
|
693 |
-
if __name__ == "__main__":
|
694 |
-
demo.queue().launch()
|
|
|
1 |
+
# Copyright (c) Meta Platforms, Inc. and affiliates
|
2 |
+
# All rights reserved.
|
3 |
+
#
|
4 |
+
# This source code is licensed under the license found in the
|
5 |
+
# LICENSE file in the root directory of this source tree.
|
6 |
+
|
7 |
+
from __future__ import annotations
|
8 |
+
|
9 |
+
import gradio as gr
|
10 |
+
import numpy as np
|
11 |
+
# import torch
|
12 |
+
|
13 |
+
|
14 |
+
from gradio_client import Client
|
15 |
+
|
16 |
+
client = Client("https://facebook-seamless-m4t.hf.space/")
|
17 |
+
|
18 |
+
DESCRIPTION = """
|
19 |
+
|
20 |
+
# SM4T
|
21 |
+
|
22 |
+
Ứng dụng có thể chuyển đổi giọng nói hoặc chữ viết sang giọng nói hoặc chữ viết của một ngôn ngữ khác.
|
23 |
+
\nHiện tại SM4T đã hỗ trợ 94 ngôn ngữ khác nhau.
|
24 |
+
|
25 |
+
"""
|
26 |
+
|
27 |
+
TASK_NAMES = [
|
28 |
+
"S2ST (Speech to Speech translation)",
|
29 |
+
"S2TT (Speech to Text translation)",
|
30 |
+
"T2ST (Text to Speech translation)",
|
31 |
+
"T2TT (Text to Text translation)",
|
32 |
+
"ASR (Automatic Speech Recognition)",
|
33 |
+
]
|
34 |
+
|
35 |
+
# Language dict
|
36 |
+
language_code_to_name = {
|
37 |
+
"afr": "Afrikaans",
|
38 |
+
"amh": "Amharic",
|
39 |
+
"arb": "Modern Standard Arabic",
|
40 |
+
"ary": "Moroccan Arabic",
|
41 |
+
"arz": "Egyptian Arabic",
|
42 |
+
"asm": "Assamese",
|
43 |
+
"ast": "Asturian",
|
44 |
+
"azj": "North Azerbaijani",
|
45 |
+
"bel": "Belarusian",
|
46 |
+
"ben": "Bengali",
|
47 |
+
"bos": "Bosnian",
|
48 |
+
"bul": "Bulgarian",
|
49 |
+
"cat": "Catalan",
|
50 |
+
"ceb": "Cebuano",
|
51 |
+
"ces": "Czech",
|
52 |
+
"ckb": "Central Kurdish",
|
53 |
+
"cmn": "Mandarin Chinese",
|
54 |
+
"cym": "Welsh",
|
55 |
+
"dan": "Danish",
|
56 |
+
"deu": "German",
|
57 |
+
"ell": "Greek",
|
58 |
+
"eng": "English",
|
59 |
+
"est": "Estonian",
|
60 |
+
"eus": "Basque",
|
61 |
+
"fin": "Finnish",
|
62 |
+
"fra": "French",
|
63 |
+
"gaz": "West Central Oromo",
|
64 |
+
"gle": "Irish",
|
65 |
+
"glg": "Galician",
|
66 |
+
"guj": "Gujarati",
|
67 |
+
"heb": "Hebrew",
|
68 |
+
"hin": "Hindi",
|
69 |
+
"hrv": "Croatian",
|
70 |
+
"hun": "Hungarian",
|
71 |
+
"hye": "Armenian",
|
72 |
+
"ibo": "Igbo",
|
73 |
+
"ind": "Indonesian",
|
74 |
+
"isl": "Icelandic",
|
75 |
+
"ita": "Italian",
|
76 |
+
"jav": "Javanese",
|
77 |
+
"jpn": "Japanese",
|
78 |
+
"kam": "Kamba",
|
79 |
+
"kan": "Kannada",
|
80 |
+
"kat": "Georgian",
|
81 |
+
"kaz": "Kazakh",
|
82 |
+
"kea": "Kabuverdianu",
|
83 |
+
"khk": "Halh Mongolian",
|
84 |
+
"khm": "Khmer",
|
85 |
+
"kir": "Kyrgyz",
|
86 |
+
"kor": "Korean",
|
87 |
+
"lao": "Lao",
|
88 |
+
"lit": "Lithuanian",
|
89 |
+
"ltz": "Luxembourgish",
|
90 |
+
"lug": "Ganda",
|
91 |
+
"luo": "Luo",
|
92 |
+
"lvs": "Standard Latvian",
|
93 |
+
"mai": "Maithili",
|
94 |
+
"mal": "Malayalam",
|
95 |
+
"mar": "Marathi",
|
96 |
+
"mkd": "Macedonian",
|
97 |
+
"mlt": "Maltese",
|
98 |
+
"mni": "Meitei",
|
99 |
+
"mya": "Burmese",
|
100 |
+
"nld": "Dutch",
|
101 |
+
"nno": "Norwegian Nynorsk",
|
102 |
+
"nob": "Norwegian Bokm\u00e5l",
|
103 |
+
"npi": "Nepali",
|
104 |
+
"nya": "Nyanja",
|
105 |
+
"oci": "Occitan",
|
106 |
+
"ory": "Odia",
|
107 |
+
"pan": "Punjabi",
|
108 |
+
"pbt": "Southern Pashto",
|
109 |
+
"pes": "Western Persian",
|
110 |
+
"pol": "Polish",
|
111 |
+
"por": "Portuguese",
|
112 |
+
"ron": "Romanian",
|
113 |
+
"rus": "Russian",
|
114 |
+
"slk": "Slovak",
|
115 |
+
"slv": "Slovenian",
|
116 |
+
"sna": "Shona",
|
117 |
+
"snd": "Sindhi",
|
118 |
+
"som": "Somali",
|
119 |
+
"spa": "Spanish",
|
120 |
+
"srp": "Serbian",
|
121 |
+
"swe": "Swedish",
|
122 |
+
"swh": "Swahili",
|
123 |
+
"tam": "Tamil",
|
124 |
+
"tel": "Telugu",
|
125 |
+
"tgk": "Tajik",
|
126 |
+
"tgl": "Tagalog",
|
127 |
+
"tha": "Thai",
|
128 |
+
"tur": "Turkish",
|
129 |
+
"ukr": "Ukrainian",
|
130 |
+
"urd": "Urdu",
|
131 |
+
"uzn": "Northern Uzbek",
|
132 |
+
"vie": "Vietnamese",
|
133 |
+
"xho": "Xhosa",
|
134 |
+
"yor": "Yoruba",
|
135 |
+
"yue": "Cantonese",
|
136 |
+
"zlm": "Colloquial Malay",
|
137 |
+
"zsm": "Standard Malay",
|
138 |
+
"zul": "Zulu",
|
139 |
+
}
|
140 |
+
LANGUAGE_NAME_TO_CODE = {v: k for k, v in language_code_to_name.items()}
|
141 |
+
|
142 |
+
# Source langs: S2ST / S2TT / ASR don't need source lang
|
143 |
+
# T2TT / T2ST use this
|
144 |
+
text_source_language_codes = [
|
145 |
+
"afr",
|
146 |
+
"amh",
|
147 |
+
"arb",
|
148 |
+
"ary",
|
149 |
+
"arz",
|
150 |
+
"asm",
|
151 |
+
"azj",
|
152 |
+
"bel",
|
153 |
+
"ben",
|
154 |
+
"bos",
|
155 |
+
"bul",
|
156 |
+
"cat",
|
157 |
+
"ceb",
|
158 |
+
"ces",
|
159 |
+
"ckb",
|
160 |
+
"cmn",
|
161 |
+
"cym",
|
162 |
+
"dan",
|
163 |
+
"deu",
|
164 |
+
"ell",
|
165 |
+
"eng",
|
166 |
+
"est",
|
167 |
+
"eus",
|
168 |
+
"fin",
|
169 |
+
"fra",
|
170 |
+
"gaz",
|
171 |
+
"gle",
|
172 |
+
"glg",
|
173 |
+
"guj",
|
174 |
+
"heb",
|
175 |
+
"hin",
|
176 |
+
"hrv",
|
177 |
+
"hun",
|
178 |
+
"hye",
|
179 |
+
"ibo",
|
180 |
+
"ind",
|
181 |
+
"isl",
|
182 |
+
"ita",
|
183 |
+
"jav",
|
184 |
+
"jpn",
|
185 |
+
"kan",
|
186 |
+
"kat",
|
187 |
+
"kaz",
|
188 |
+
"khk",
|
189 |
+
"khm",
|
190 |
+
"kir",
|
191 |
+
"kor",
|
192 |
+
"lao",
|
193 |
+
"lit",
|
194 |
+
"lug",
|
195 |
+
"luo",
|
196 |
+
"lvs",
|
197 |
+
"mai",
|
198 |
+
"mal",
|
199 |
+
"mar",
|
200 |
+
"mkd",
|
201 |
+
"mlt",
|
202 |
+
"mni",
|
203 |
+
"mya",
|
204 |
+
"nld",
|
205 |
+
"nno",
|
206 |
+
"nob",
|
207 |
+
"npi",
|
208 |
+
"nya",
|
209 |
+
"ory",
|
210 |
+
"pan",
|
211 |
+
"pbt",
|
212 |
+
"pes",
|
213 |
+
"pol",
|
214 |
+
"por",
|
215 |
+
"ron",
|
216 |
+
"rus",
|
217 |
+
"slk",
|
218 |
+
"slv",
|
219 |
+
"sna",
|
220 |
+
"snd",
|
221 |
+
"som",
|
222 |
+
"spa",
|
223 |
+
"srp",
|
224 |
+
"swe",
|
225 |
+
"swh",
|
226 |
+
"tam",
|
227 |
+
"tel",
|
228 |
+
"tgk",
|
229 |
+
"tgl",
|
230 |
+
"tha",
|
231 |
+
"tur",
|
232 |
+
"ukr",
|
233 |
+
"urd",
|
234 |
+
"uzn",
|
235 |
+
"vie",
|
236 |
+
"yor",
|
237 |
+
"yue",
|
238 |
+
"zsm",
|
239 |
+
"zul",
|
240 |
+
]
|
241 |
+
TEXT_SOURCE_LANGUAGE_NAMES = sorted(
|
242 |
+
[language_code_to_name[code] for code in text_source_language_codes]
|
243 |
+
)
|
244 |
+
|
245 |
+
# Target langs:
|
246 |
+
# S2ST / T2ST
|
247 |
+
s2st_target_language_codes = [
|
248 |
+
"eng",
|
249 |
+
"arb",
|
250 |
+
"ben",
|
251 |
+
"cat",
|
252 |
+
"ces",
|
253 |
+
"cmn",
|
254 |
+
"cym",
|
255 |
+
"dan",
|
256 |
+
"deu",
|
257 |
+
"est",
|
258 |
+
"fin",
|
259 |
+
"fra",
|
260 |
+
"hin",
|
261 |
+
"ind",
|
262 |
+
"ita",
|
263 |
+
"jpn",
|
264 |
+
"kor",
|
265 |
+
"mlt",
|
266 |
+
"nld",
|
267 |
+
"pes",
|
268 |
+
"pol",
|
269 |
+
"por",
|
270 |
+
"ron",
|
271 |
+
"rus",
|
272 |
+
"slk",
|
273 |
+
"spa",
|
274 |
+
"swe",
|
275 |
+
"swh",
|
276 |
+
"tel",
|
277 |
+
"tgl",
|
278 |
+
"tha",
|
279 |
+
"tur",
|
280 |
+
"ukr",
|
281 |
+
"urd",
|
282 |
+
"uzn",
|
283 |
+
"vie",
|
284 |
+
]
|
285 |
+
S2ST_TARGET_LANGUAGE_NAMES = sorted(
|
286 |
+
[language_code_to_name[code] for code in s2st_target_language_codes]
|
287 |
+
)
|
288 |
+
# S2TT / ASR
|
289 |
+
S2TT_TARGET_LANGUAGE_NAMES = TEXT_SOURCE_LANGUAGE_NAMES
|
290 |
+
# T2TT
|
291 |
+
T2TT_TARGET_LANGUAGE_NAMES = TEXT_SOURCE_LANGUAGE_NAMES
|
292 |
+
|
293 |
+
# Download sample input audio files
|
294 |
+
filenames = ["assets/sample_input.mp3", "assets/sample_input_2.mp3"]
|
295 |
+
# for filename in filenames:
|
296 |
+
# hf_hub_download(
|
297 |
+
# repo_id="facebook/seamless_m4t",
|
298 |
+
# repo_type="space",
|
299 |
+
# filename=filename,
|
300 |
+
# local_dir=".",
|
301 |
+
# )
|
302 |
+
|
303 |
+
AUDIO_SAMPLE_RATE = 16000.0
|
304 |
+
MAX_INPUT_AUDIO_LENGTH = 60 # in seconds
|
305 |
+
DEFAULT_TARGET_LANGUAGE = "French"
|
306 |
+
|
307 |
+
# device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
308 |
+
|
309 |
+
def api_predict(
|
310 |
+
task_name: str,
|
311 |
+
audio_source: str,
|
312 |
+
input_audio_mic: str | None,
|
313 |
+
input_audio_file: str | None,
|
314 |
+
input_text: str | None,
|
315 |
+
source_language: str | None,
|
316 |
+
target_language: str,):
|
317 |
+
|
318 |
+
audio_out, text_out = client.predict(task_name,
|
319 |
+
audio_source,
|
320 |
+
input_audio_mic,
|
321 |
+
input_audio_file,
|
322 |
+
input_text,
|
323 |
+
source_language,
|
324 |
+
target_language,
|
325 |
+
api_name="/run")
|
326 |
+
return audio_out, text_out
|
327 |
+
|
328 |
+
|
329 |
+
|
330 |
+
|
331 |
+
|
332 |
+
def process_s2st_example(
|
333 |
+
input_audio_file: str, target_language: str
|
334 |
+
) -> tuple[tuple[int, np.ndarray] | None, str]:
|
335 |
+
return api_predict(
|
336 |
+
task_name="S2ST",
|
337 |
+
audio_source="file",
|
338 |
+
input_audio_mic=None,
|
339 |
+
input_audio_file=input_audio_file,
|
340 |
+
input_text=None,
|
341 |
+
source_language=None,
|
342 |
+
target_language=target_language,
|
343 |
+
)
|
344 |
+
|
345 |
+
|
346 |
+
def process_s2tt_example(
|
347 |
+
input_audio_file: str, target_language: str
|
348 |
+
) -> tuple[tuple[int, np.ndarray] | None, str]:
|
349 |
+
return api_predict(
|
350 |
+
task_name="S2TT",
|
351 |
+
audio_source="file",
|
352 |
+
input_audio_mic=None,
|
353 |
+
input_audio_file=input_audio_file,
|
354 |
+
input_text=None,
|
355 |
+
source_language=None,
|
356 |
+
target_language=target_language,
|
357 |
+
)
|
358 |
+
|
359 |
+
|
360 |
+
def process_t2st_example(
|
361 |
+
input_text: str, source_language: str, target_language: str
|
362 |
+
) -> tuple[tuple[int, np.ndarray] | None, str]:
|
363 |
+
return api_predict(
|
364 |
+
task_name="T2ST",
|
365 |
+
audio_source="",
|
366 |
+
input_audio_mic=None,
|
367 |
+
input_audio_file=None,
|
368 |
+
input_text=input_text,
|
369 |
+
source_language=source_language,
|
370 |
+
target_language=target_language,
|
371 |
+
)
|
372 |
+
|
373 |
+
|
374 |
+
def process_t2tt_example(
|
375 |
+
input_text: str, source_language: str, target_language: str
|
376 |
+
) -> tuple[tuple[int, np.ndarray] | None, str]:
|
377 |
+
return api_predict(
|
378 |
+
task_name="T2TT",
|
379 |
+
audio_source="",
|
380 |
+
input_audio_mic=None,
|
381 |
+
input_audio_file=None,
|
382 |
+
input_text=input_text,
|
383 |
+
source_language=source_language,
|
384 |
+
target_language=target_language,
|
385 |
+
)
|
386 |
+
|
387 |
+
|
388 |
+
def process_asr_example(
|
389 |
+
input_audio_file: str, target_language: str
|
390 |
+
) -> tuple[tuple[int, np.ndarray] | None, str]:
|
391 |
+
return api_predict(
|
392 |
+
task_name="ASR",
|
393 |
+
audio_source="file",
|
394 |
+
input_audio_mic=None,
|
395 |
+
input_audio_file=input_audio_file,
|
396 |
+
input_text=None,
|
397 |
+
source_language=None,
|
398 |
+
target_language=target_language,
|
399 |
+
)
|
400 |
+
|
401 |
+
|
402 |
+
def update_audio_ui(audio_source: str) -> tuple[dict, dict]:
|
403 |
+
mic = audio_source == "microphone"
|
404 |
+
return (
|
405 |
+
gr.update(visible=mic, value=None), # input_audio_mic
|
406 |
+
gr.update(visible=not mic, value=None), # input_audio_file
|
407 |
+
)
|
408 |
+
|
409 |
+
|
410 |
+
def update_input_ui(task_name: str) -> tuple[dict, dict, dict, dict]:
|
411 |
+
task_name = task_name.split()[0]
|
412 |
+
if task_name == "S2ST":
|
413 |
+
return (
|
414 |
+
gr.update(visible=True), # audio_box
|
415 |
+
gr.update(visible=False), # input_text
|
416 |
+
gr.update(visible=False), # source_language
|
417 |
+
gr.update(
|
418 |
+
visible=True,
|
419 |
+
choices=S2ST_TARGET_LANGUAGE_NAMES,
|
420 |
+
value=DEFAULT_TARGET_LANGUAGE,
|
421 |
+
), # target_language
|
422 |
+
)
|
423 |
+
elif task_name == "S2TT":
|
424 |
+
return (
|
425 |
+
gr.update(visible=True), # audio_box
|
426 |
+
gr.update(visible=False), # input_text
|
427 |
+
gr.update(visible=False), # source_language
|
428 |
+
gr.update(
|
429 |
+
visible=True,
|
430 |
+
choices=S2TT_TARGET_LANGUAGE_NAMES,
|
431 |
+
value=DEFAULT_TARGET_LANGUAGE,
|
432 |
+
), # target_language
|
433 |
+
)
|
434 |
+
elif task_name == "T2ST":
|
435 |
+
return (
|
436 |
+
gr.update(visible=False), # audio_box
|
437 |
+
gr.update(visible=True), # input_text
|
438 |
+
gr.update(visible=True), # source_language
|
439 |
+
gr.update(
|
440 |
+
visible=True,
|
441 |
+
choices=S2ST_TARGET_LANGUAGE_NAMES,
|
442 |
+
value=DEFAULT_TARGET_LANGUAGE,
|
443 |
+
), # target_language
|
444 |
+
)
|
445 |
+
elif task_name == "T2TT":
|
446 |
+
return (
|
447 |
+
gr.update(visible=False), # audio_box
|
448 |
+
gr.update(visible=True), # input_text
|
449 |
+
gr.update(visible=True), # source_language
|
450 |
+
gr.update(
|
451 |
+
visible=True,
|
452 |
+
choices=T2TT_TARGET_LANGUAGE_NAMES,
|
453 |
+
value=DEFAULT_TARGET_LANGUAGE,
|
454 |
+
), # target_language
|
455 |
+
)
|
456 |
+
elif task_name == "ASR":
|
457 |
+
return (
|
458 |
+
gr.update(visible=True), # audio_box
|
459 |
+
gr.update(visible=False), # input_text
|
460 |
+
gr.update(visible=False), # source_language
|
461 |
+
gr.update(
|
462 |
+
visible=True,
|
463 |
+
choices=S2TT_TARGET_LANGUAGE_NAMES,
|
464 |
+
value=DEFAULT_TARGET_LANGUAGE,
|
465 |
+
), # target_language
|
466 |
+
)
|
467 |
+
else:
|
468 |
+
raise ValueError(f"Unknown task: {task_name}")
|
469 |
+
|
470 |
+
|
471 |
+
def update_output_ui(task_name: str) -> tuple[dict, dict]:
|
472 |
+
task_name = task_name.split()[0]
|
473 |
+
if task_name in ["S2ST", "T2ST"]:
|
474 |
+
return (
|
475 |
+
gr.update(visible=True, value=None), # output_audio
|
476 |
+
gr.update(value=None), # output_text
|
477 |
+
)
|
478 |
+
elif task_name in ["S2TT", "T2TT", "ASR"]:
|
479 |
+
return (
|
480 |
+
gr.update(visible=False, value=None), # output_audio
|
481 |
+
gr.update(value=None), # output_text
|
482 |
+
)
|
483 |
+
else:
|
484 |
+
raise ValueError(f"Unknown task: {task_name}")
|
485 |
+
|
486 |
+
|
487 |
+
def update_example_ui(task_name: str) -> tuple[dict, dict, dict, dict, dict]:
|
488 |
+
task_name = task_name.split()[0]
|
489 |
+
return (
|
490 |
+
gr.update(visible=task_name == "S2ST"), # s2st_example_row
|
491 |
+
gr.update(visible=task_name == "S2TT"), # s2tt_example_row
|
492 |
+
gr.update(visible=task_name == "T2ST"), # t2st_example_row
|
493 |
+
gr.update(visible=task_name == "T2TT"), # t2tt_example_row
|
494 |
+
gr.update(visible=task_name == "ASR"), # asr_example_row
|
495 |
+
)
|
496 |
+
|
497 |
+
|
498 |
+
css = """
|
499 |
+
h1 {
|
500 |
+
text-align: center;
|
501 |
+
}
|
502 |
+
|
503 |
+
#.contain {
|
504 |
+
# max-width: 730px;
|
505 |
+
# margin: auto;
|
506 |
+
# padding-top: 1.5rem;
|
507 |
+
#}
|
508 |
+
"""
|
509 |
+
|
510 |
+
with gr.Blocks(css=css) as demo:
|
511 |
+
gr.Markdown(DESCRIPTION)
|
512 |
+
with gr.Group():
|
513 |
+
task_name = gr.Dropdown(
|
514 |
+
label="Task",
|
515 |
+
choices=TASK_NAMES,
|
516 |
+
value=TASK_NAMES[0],
|
517 |
+
)
|
518 |
+
with gr.Row():
|
519 |
+
source_language = gr.Dropdown(
|
520 |
+
label="Source language",
|
521 |
+
choices=TEXT_SOURCE_LANGUAGE_NAMES,
|
522 |
+
value="English",
|
523 |
+
visible=False,
|
524 |
+
)
|
525 |
+
target_language = gr.Dropdown(
|
526 |
+
label="Target language",
|
527 |
+
choices=S2ST_TARGET_LANGUAGE_NAMES,
|
528 |
+
value=DEFAULT_TARGET_LANGUAGE,
|
529 |
+
)
|
530 |
+
with gr.Row() as audio_box:
|
531 |
+
audio_source = gr.Radio(
|
532 |
+
label="Audio source",
|
533 |
+
choices=["file", "microphone"],
|
534 |
+
value="file",
|
535 |
+
)
|
536 |
+
input_audio_mic = gr.Audio(
|
537 |
+
label="Input speech",
|
538 |
+
type="filepath",
|
539 |
+
source="microphone",
|
540 |
+
visible=False,
|
541 |
+
)
|
542 |
+
input_audio_file = gr.Audio(
|
543 |
+
label="Input speech",
|
544 |
+
type="filepath",
|
545 |
+
source="upload",
|
546 |
+
visible=True,
|
547 |
+
)
|
548 |
+
input_text = gr.Textbox(label="Input text", visible=False)
|
549 |
+
with gr.Row():
|
550 |
+
btn = gr.Button("Translate")
|
551 |
+
btn_clean = gr.ClearButton([input_audio_mic, input_audio_file])
|
552 |
+
# gr.Markdown("## Text Examples")
|
553 |
+
with gr.Column():
|
554 |
+
output_audio = gr.Audio(
|
555 |
+
label="Translated speech",
|
556 |
+
autoplay=False,
|
557 |
+
streaming=False,
|
558 |
+
type="numpy",
|
559 |
+
)
|
560 |
+
output_text = gr.Textbox(label="Translated text")
|
561 |
+
|
562 |
+
with gr.Row(visible=True) as s2st_example_row:
|
563 |
+
s2st_examples = gr.Examples(
|
564 |
+
examples=[
|
565 |
+
["assets/sample_input.mp3", "French"],
|
566 |
+
["assets/sample_input.mp3", "Mandarin Chinese"],
|
567 |
+
["assets/sample_input_2.mp3", "Hindi"],
|
568 |
+
["assets/sample_input_2.mp3", "Spanish"],
|
569 |
+
],
|
570 |
+
inputs=[input_audio_file, target_language],
|
571 |
+
outputs=[output_audio, output_text],
|
572 |
+
fn=process_s2st_example,
|
573 |
+
)
|
574 |
+
with gr.Row(visible=False) as s2tt_example_row:
|
575 |
+
s2tt_examples = gr.Examples(
|
576 |
+
examples=[
|
577 |
+
["assets/sample_input.mp3", "French"],
|
578 |
+
["assets/sample_input.mp3", "Mandarin Chinese"],
|
579 |
+
["assets/sample_input_2.mp3", "Hindi"],
|
580 |
+
["assets/sample_input_2.mp3", "Spanish"],
|
581 |
+
],
|
582 |
+
inputs=[input_audio_file, target_language],
|
583 |
+
outputs=[output_audio, output_text],
|
584 |
+
fn=process_s2tt_example,
|
585 |
+
)
|
586 |
+
with gr.Row(visible=False) as t2st_example_row:
|
587 |
+
t2st_examples = gr.Examples(
|
588 |
+
examples=[
|
589 |
+
["My favorite animal is the elephant.", "English", "French"],
|
590 |
+
["My favorite animal is the elephant.", "English", "Mandarin Chinese"],
|
591 |
+
[
|
592 |
+
"Meta AI's Seamless M4T model is democratising spoken communication across language barriers",
|
593 |
+
"English",
|
594 |
+
"Hindi",
|
595 |
+
],
|
596 |
+
[
|
597 |
+
"Meta AI's Seamless M4T model is democratising spoken communication across language barriers",
|
598 |
+
"English",
|
599 |
+
"Spanish",
|
600 |
+
],
|
601 |
+
],
|
602 |
+
inputs=[input_text, source_language, target_language],
|
603 |
+
outputs=[output_audio, output_text],
|
604 |
+
fn=process_t2st_example,
|
605 |
+
)
|
606 |
+
with gr.Row(visible=False) as t2tt_example_row:
|
607 |
+
t2tt_examples = gr.Examples(
|
608 |
+
examples=[
|
609 |
+
["My favorite animal is the elephant.", "English", "French"],
|
610 |
+
["My favorite animal is the elephant.", "English", "Mandarin Chinese"],
|
611 |
+
[
|
612 |
+
"Meta AI's Seamless M4T model is democratising spoken communication across language barriers",
|
613 |
+
"English",
|
614 |
+
"Hindi",
|
615 |
+
],
|
616 |
+
[
|
617 |
+
"Meta AI's Seamless M4T model is democratising spoken communication across language barriers",
|
618 |
+
"English",
|
619 |
+
"Spanish",
|
620 |
+
],
|
621 |
+
],
|
622 |
+
inputs=[input_text, source_language, target_language],
|
623 |
+
outputs=[output_audio, output_text],
|
624 |
+
fn=process_t2tt_example,
|
625 |
+
)
|
626 |
+
with gr.Row(visible=False) as asr_example_row:
|
627 |
+
asr_examples = gr.Examples(
|
628 |
+
examples=[
|
629 |
+
["assets/sample_input.mp3", "English"],
|
630 |
+
["assets/sample_input_2.mp3", "English"],
|
631 |
+
],
|
632 |
+
inputs=[input_audio_file, target_language],
|
633 |
+
outputs=[output_audio, output_text],
|
634 |
+
fn=process_asr_example,
|
635 |
+
)
|
636 |
+
|
637 |
+
audio_source.change(
|
638 |
+
fn=update_audio_ui,
|
639 |
+
inputs=audio_source,
|
640 |
+
outputs=[
|
641 |
+
input_audio_mic,
|
642 |
+
input_audio_file,
|
643 |
+
],
|
644 |
+
queue=False,
|
645 |
+
api_name=False,
|
646 |
+
)
|
647 |
+
task_name.change(
|
648 |
+
fn=update_input_ui,
|
649 |
+
inputs=task_name,
|
650 |
+
outputs=[
|
651 |
+
audio_box,
|
652 |
+
input_text,
|
653 |
+
source_language,
|
654 |
+
target_language,
|
655 |
+
],
|
656 |
+
queue=False,
|
657 |
+
api_name=False,
|
658 |
+
).then(
|
659 |
+
fn=update_output_ui,
|
660 |
+
inputs=task_name,
|
661 |
+
outputs=[output_audio, output_text],
|
662 |
+
queue=False,
|
663 |
+
api_name=False,
|
664 |
+
).then(
|
665 |
+
fn=update_example_ui,
|
666 |
+
inputs=task_name,
|
667 |
+
outputs=[
|
668 |
+
s2st_example_row,
|
669 |
+
s2tt_example_row,
|
670 |
+
t2st_example_row,
|
671 |
+
t2tt_example_row,
|
672 |
+
asr_example_row,
|
673 |
+
],
|
674 |
+
queue=False,
|
675 |
+
api_name=False,
|
676 |
+
)
|
677 |
+
|
678 |
+
btn.click(
|
679 |
+
fn=api_predict,
|
680 |
+
inputs=[
|
681 |
+
task_name,
|
682 |
+
audio_source,
|
683 |
+
input_audio_mic,
|
684 |
+
input_audio_file,
|
685 |
+
input_text,
|
686 |
+
source_language,
|
687 |
+
target_language,
|
688 |
+
],
|
689 |
+
outputs=[output_audio, output_text],
|
690 |
+
api_name="run",
|
691 |
+
)
|
692 |
+
|
693 |
+
if __name__ == "__main__":
|
694 |
+
demo.queue().launch()
|