File size: 14,354 Bytes
222e3bd
e935ff6
 
ea6a933
83ebf46
c79df46
2ddb634
e28791b
222e3bd
2ad24b2
c79df46
6f94cd7
 
 
 
 
 
 
 
f2cf91b
e494712
 
f2cf91b
a0a523d
 
 
 
 
f2cf91b
 
0191adb
10cf936
0191adb
 
 
10cf936
 
0191adb
10cf936
 
f2cf91b
 
044fea4
f2cf91b
0191adb
3c34d6c
76e9cbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3c34d6c
e28791b
0191adb
a1d10f4
0191adb
 
a1d10f4
0191adb
 
a1d10f4
0191adb
a1d10f4
 
 
0191adb
 
 
a1d10f4
0191adb
 
a1d10f4
 
0191adb
 
a1d10f4
0191adb
 
 
 
a1d10f4
0191adb
 
a1d10f4
 
2ddb634
e1a50fc
 
e283de7
2ddb634
9dcf11b
 
f67cac3
9dcf11b
 
 
f67cac3
 
7d90a74
37132c3
3980780
 
 
dd3c056
3980780
 
0e7dc9e
f67cac3
 
2a8b22c
0191adb
2a8b22c
2ddb634
 
4ea5474
9366d4b
 
2e101cc
5c4653b
 
 
 
 
 
f2cf91b
5c4653b
 
 
 
a65b632
044fea4
 
22b65d9
a65b632
2ad24b2
 
 
9366d4b
044fea4
 
 
 
a65b632
22b65d9
9366d4b
044fea4
5c4653b
83ebf46
 
 
 
 
 
 
 
 
 
6115840
 
83ebf46
e1c08c5
 
1cd6967
f2cf91b
9366d4b
044fea4
1cd6967
ea6a933
222e3bd
6299b6a
3c34d6c
ea6a933
 
 
83ebf46
a4ca82b
 
 
 
6115840
 
a4ca82b
83ebf46
6115840
ea6a933
83ebf46
2ddb634
81f118e
 
2ddb634
 
 
 
 
 
 
ea6a933
3980780
a0a523d
3980780
2ad24b2
 
 
 
3980780
22b65d9
a0a523d
f4f5edf
1ae721a
 
 
 
 
 
 
 
dcbacca
5c1072f
 
 
76e9cbc
f4f5edf
76e9cbc
1ae721a
76e9cbc
1ae721a
76e9cbc
 
 
f4f5edf
 
1ae721a
 
 
f4f5edf
1ae721a
f4f5edf
1ae721a
76e9cbc
 
f4f5edf
76e9cbc
 
5c1072f
76e9cbc
 
 
 
f4f5edf
 
76e9cbc
4ea5474
 
2ddb634
a1d10f4
0191adb
2ddb634
79e5c13
2ddb634
4ea5474
a1d10f4
83ebf46
 
0191adb
 
 
 
9366d4b
 
0191adb
9366d4b
0191adb
e925b78
a1d10f4
76e9cbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a1d10f4
3c34d6c
 
b1e470d
3c34d6c
a1d10f4
3c34d6c
4ea5474
76e9cbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
22b65d9
 
 
 
 
 
76e9cbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f4f5edf
 
 
76e9cbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2ddb634
4ea5474
2ddb634
675fe4f
2ad24b2
 
 
 
 
 
 
 
 
2ddb634
63b454b
76e9cbc
2ddb634
 
2ad24b2
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
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
import os
import sys
import time
import requests
import json
from subprocess import Popen, PIPE
import threading
from huggingface_hub import hf_hub_download
import gradio as gr
import resources.app.no_server as xvaserver

try:
	import resources.app.no_server as xvaserver
except:
	print('server.log contents:')
	with open('resources/app/server.log', 'r') as f:
		print(f.read())


hf_model_name = "Pendrokar/xvapitch_nvidia"
hf_cache_models_path = '/home/user/.cache/huggingface/hub/models--Pendrokar--xvapitch_nvidia/snapshots/61b10e60b22bc21c1e072f72f1108b9c2b21e94c/'
models_path = '/home/user/.cache/huggingface/hub/models--Pendrokar--xvapitch_nvidia/snapshots/61b10e60b22bc21c1e072f72f1108b9c2b21e94c/'

# FIXME: currently hardcoded in DeepMoji code
# try:
# 	os.symlink('/home/user/.cache/huggingface/hub/models--Pendrokar--TorchMoji/snapshots/58217568daaf64d3621245dd5c88c94e651a08d6', '/home/user/app/resources/app/plugins/deepmoji_plugings/model', target_is_directory=True)
# except:
# 	print('Failed to create symlink to DeepMoji model, may already be there.')

voice_models = [
	("Male #6671", "ccby_nvidia_hifi_6671_M"),
	("Male #6670", "ccby_nvidia_hifi_6670_M"),
	("Male #9017", "ccby_nvidia_hifi_9017_M"),
	("Male #6097", "ccby_nvidia_hifi_6097_M"),
	("Female #92", "ccby_nvidia_hifi_92_F"),
	("Female #11697", "ccby_nvidia_hifi_11697_F"),
	("Female #12787", "ccby_nvidia_hifi_12787_F"),
	("Female #11614", "ccby_nv_hifi_11614_F"),
	("Female #8051", "ccby_nvidia_hifi_8051_F"),
	("Female #9136", "ccby_nvidia_hifi_9136_F"),
]
current_voice_model = None
base_speaker_emb = ''

# order ranked by similarity to English due to the xVASynth's use of ARPAbet instead of IPA
languages = [
	("🇬🇧 EN", "en"),
	("🇩🇪 DE", "de"),
	("🇪🇸 ES", "es"),
	("🇮🇹 IT", "it"),
	("🇳🇱 NL", "nl"),
	("🇵🇹 PT", "pt"),
	("🇵🇱 PL", "pl"),
	("🇷🇴 RO", "ro"),
	("🇸🇪 SV", "sv"),
	("🇩🇰 DA", "da"),
	("🇫🇮 FI", "fi"),
	("🇭🇺 HU", "hu"),
	("🇬🇷 EL", "el"),
	("🇫🇷 FR", "fr"),
	("🇷🇺 RU", "ru"),
	("🇺🇦 UK", "uk"),
	("🇹🇷 TR", "tr"),
	("🇸🇦 AR", "ar"),
	("🇮🇳 HI", "hi"),
	("🇯🇵 JP", "jp"),
	("🇰🇷 KO", "ko"),
	("🇨🇳 ZH", "zh"),
	("🇻🇳 VI", "vi"),
	("🇻🇦 LA", "la"),
	("HA", "ha"),
	("SW", "sw"),
	("🇳🇬 YO", "yo"),
	("WO", "wo"),
]

# Translated from English by DeepMind's Gemini Pro
default_text = {
	"ar": "هذا هو صوتي.",
	"da": "Sådan lyder min stemme.",
	"de": "So klingt meine Stimme.",
	"el": "Έτσι ακούγεται η φωνή μου.",
	"en": "This is what my voice sounds like.",
	"es": "Así suena mi voz.",
	"fi": "Näin ääneni kuulostaa.",
	"fr": "Voici à quoi ressemble ma voix.",
	"ha": "Wannan ne muryata ke.",
	"hi": "यह मेरी आवाज़ कैसी लगती है।",
	"hu": "Így hangzik a hangom.",
	"it": "Così suona la mia voce.",
	"jp": "これが私の声です。",
	"ko": "여기 제 목소리가 어떤지 들어보세요.",
	"la": "Haec est vox mea sonans.",
	"nl": "Dit is hoe mijn stem klinkt.",
	"pl": "Tak brzmi mój głos.",
	"pt": "É assim que minha voz soa.",
	"ro": "Așa sună vocea mea.",
	"ru": "Вот как звучит мой голос.",
	"sv": "Såhär låter min röst.",
	"sw": "Sauti yangu inasikika hivi.",
	"tr": "Benim sesimin sesi böyle.",
	"uk": "Ось як звучить мій голос.",
	"vi": "Đây là giọng nói của tôi.",
	"wo": "Ndox li neen xewnaal ma.",
	"yo": "Ìyí ni ohùn mi ńlá.",
	"zh": "这是我的声音。",
}

def run_xvaserver():
	# start the process without waiting for a response
	print('Running xVAServer subprocess...\n')
	xvaserver = Popen(['python', f'{os.path.dirname(os.path.abspath(__file__))}/resources/app/server.py'], stdout=PIPE, stderr=PIPE, cwd=f'{os.path.dirname(os.path.abspath(__file__))}/resources/app/')

	# Wait for a moment to ensure the server starts up
	time.sleep(10)

	# Check if the server is running
	if xvaserver.poll() is not None:
		print("Web server failed to start.")
		sys.exit(0)

	# contact local xVASynth server
	print('Attempting to connect to xVASynth...')
	try:
		response = requests.get('http://0.0.0.0:8008')
		response.raise_for_status()  # If the response contains an HTTP error status code, raise an exception
	except requests.exceptions.RequestException as err:
		print('Failed to connect!')
		return

	print('xVAServer running on port 8008')

	# load default model
	load_model("ccby_nvidia_hifi_6671_M")

	# Wait for the process to exit
	xvaserver.wait()

def load_model(voice_model_name):
	model_path =  models_path + voice_model_name

	model_type = 'xVAPitch'
	language = 'en'

	data = {
		'outputs': None,
		'version': '3.0',
		'model': model_path,
		'modelType': model_type,
		'base_lang': language,
		'pluginsContext': '{}',
	}

	embs = base_speaker_emb

	print('Loading voice model...')
	try:
		xvaserver.loadModel(data)
		# response = requests.post('http://0.0.0.0:8008/loadModel', json=data, timeout=60)
		# response.raise_for_status()  # If the response contains an HTTP error status code, raise an exception
		current_voice_model = voice_model_name

		with open(model_path + '.json', 'r', encoding='utf-8') as f:
		    voice_model_json = json.load(f)
		embs = voice_model_json['games'][0]['base_speaker_emb']
	except requests.exceptions.RequestException as err:
		print(f'FAILED to load voice model: {err}')

	return embs

def predict(
	input_text,
	voice,
	lang,
	pacing,
	pitch,
	energy,
	anger,
	happy,
	sad,
	surprise,
	use_deepmoji
):
	# grab only the first 1000 characters
	input_text = input_text[:1000]

	# load voice model if not the current model
	if (current_voice_model != voice):
		base_speaker_emb = load_model(voice)

	model_type = 'xVAPitch'
	pace = pacing if pacing else 1.0
	save_path = '/tmp/xvapitch_audio_sample.wav'
	language = lang
	use_sr = 0
	use_cleanup = 0

	pluginsContext = {}
	pluginsContext["mantella_settings"] = {
		"emAngry": (anger if anger > 0 else 0),
		"emHappy": (happy if happy > 0 else 0),
		"emSad": (sad if sad > 0 else 0),
		"emSurprise": (surprise if surprise > 0 else 0),
		"run_model": use_deepmoji
	}


	data = {
		'pluginsContext': json.dumps(pluginsContext),
		'modelType': model_type,
		# pad with whitespaces as a workaround to avoid cutoffs
		'sequence': input_text.center(len(input_text) + 2, ' '),
		'pace': pace,
		'outfile': save_path,
		'vocoder': 'n/a',
		'base_lang': language,
		'base_emb': base_speaker_emb,
		'useSR': use_sr,
		'useCleanup': use_cleanup,
	}

	print('Synthesizing...')
	try:
		xvaserver.synthesize(data)
		# response = requests.post('http://0.0.0.0:8008/synthesize', json=data, timeout=60)
		# response.raise_for_status()  # If the response contains an HTTP error status code, raise an exception
		# json_data = json.loads(response.text)
	except requests.exceptions.RequestException as err:
		print('FAILED to synthesize: {err}')
		save_path = ''
		response = {'text': '{"message": "Failed"}'}
		json_data = {
			'arpabet': ['Failed'],
			'durations': [0],
			'em_anger': anger,
			'em_happy': happy,
			'em_sad': sad,
			'em_surprise': surprise,
		}

	print('server.log contents:')
	with open('resources/app/server.log', 'r') as f:
		print(f.read())

	arpabet_html = '<h6>ARPAbet & Phoneme lengths</h6>'
	arpabet_symbols = json_data['arpabet'].split('|')
	utter_time = 0
	for symb_i in range(len(json_data['durations'])):
		# skip PAD symbol
		if (arpabet_symbols[symb_i] == '<PAD>'):
			continue

		length = float(json_data['durations'][symb_i])
		arpa_length = str(round(length/2, 1))
		arpabet_html += '<strong\
			class="arpabet"\
			style="padding: 0 '\
			+ str(arpa_length)\
			+'em"'\
			+f" title=\"{utter_time} + {length}\""\
			+'>'\
			+ arpabet_symbols[symb_i]\
			+ '</strong> '
		utter_time += round(length, 1)

	return [
		save_path,
		arpabet_html,
		round(json_data['em_angry'][0], 2),
		round(json_data['em_happy'][0], 2),
		round(json_data['em_sad'][0], 2),
		round(json_data['em_surprise'][0], 2),
		response.text
	]

input_textbox = gr.Textbox(
	label="Input Text",
	value="This is what my voice sounds like.",
	info="Also accepts ARPAbet symbols placed within {} brackets.",
	lines=1,
	max_lines=5,
	autofocus=True
)
pacing_slider = gr.Slider(0.5, 2.0, value=1.0, step=0.1, label="Duration")
pitch_slider = gr.Slider(0, 1.0, value=0.5, step=0.05, label="Pitch", visible=False)
energy_slider = gr.Slider(0.1, 1.0, value=1.0, step=0.05, label="Energy", visible=False)
anger_slider = gr.Slider(0, 1.0, value=0, step=0.05, label="😠 Anger", info="Tread lightly beyond 0.9")
happy_slider = gr.Slider(0, 1.0, value=0, step=0.05, label="😃 Happiness", info="Tread lightly beyond 0.7")
sad_slider = gr.Slider(0, 1.0, value=0, step=0.05, label="😭 Sadness", info="Duration increased when beyond 0.2")
surprise_slider = gr.Slider(0, 1.0, value=0, step=0.05, label="😮 Surprise", info="Does not play well with Happiness with either being beyond 0.3")
voice_radio = gr.Radio(
	voice_models,
	value="ccby_nvidia_hifi_6671_M",
	label="Voice",
	info="NVIDIA HIFI CC-BY-4.0 xVAPitch voice model"
)

def set_default_text(lang, deepmoji_checked):
	# DeepMoji only works on English Text
	# checkbox_enabled = True
	# if lang != 'en':
	# 	checkbox_enabled = False

	if lang == 'en':
		checkbox_enabled = gr.Checkbox(
			label="Use DeepMoji",
			info="Auto adjust emotional values",
			value=deepmoji_checked,
			interactive=True
		)
	else:
		checkbox_enabled = gr.Checkbox(
			label="Use DeepMoji",
			info="Works only with English!",
			value=False,
			interactive=False
		)

	return default_text[lang], checkbox_enabled  # Return the modified textbox (important for Blocks)

def reset_em_sliders(
	deepmoji_enabled,
	anger,
	happy,
	sad,
	surprise
):
	if (deepmoji_enabled):
		return (0, 0, 0, 0)
	else:
		return (
			anger,
			happy,
			sad,
			surprise
		)

def toggle_deepmoji(
	checked,
	anger,
	happy,
	sad,
	surprise
):
	if checked:
		return (0, 0, 0, 0)
	else:
		return (
			anger,
			happy,
			sad,
			surprise
		)

language_radio = gr.Radio(
	languages,
	value="en",
	label="Language",
	info="Will be more monotone and have an English accent. Tested mostly by a native Briton."
)

with gr.Blocks(css=".arpabet {display: inline-block; background-color: gray; border-radius: 5px; font-size: 120%; margin: 0.1em 0}") as demo:
	gr.Markdown("# xVASynth TTS")

	with gr.Row():  # Main row for inputs and language selection
		with gr.Column():  # Input column
			input_textbox = gr.Textbox(
				label="Input Text",
				value="This is what my voice sounds like.",
				info="Also accepts ARPAbet symbols placed within {} brackets.",
				lines=1,
				max_lines=5,
				autofocus=True
			)
			language_radio = gr.Radio(
				languages,
				value="en",
				label="Language",
				info="Will be more monotone and have an English accent. Tested mostly by a native Briton."
			)
			pacing_slider = gr.Slider(0.5, 2.0, value=1.0, step=0.1, label="Duration")
		with gr.Column():  # Control column
			voice_radio = gr.Radio(
				voice_models,
				value="ccby_nvidia_hifi_6671_M",
				label="Voice",
				info="NVIDIA HIFI CC-BY-4.0 xVAPitch voice model"
			)
			pitch_slider = gr.Slider(0, 1.0, value=0.5, step=0.05, label="Pitch", visible=False)
			energy_slider = gr.Slider(0.1, 1.0, value=1.0, step=0.05, label="Energy", visible=False)
			with gr.Row():  # Main row for inputs and language selection
				with gr.Column():  # Input column
					anger_slider = gr.Slider(0, 1.0, value=0, step=0.05, label="😠 Anger", info="Tread lightly beyond 0.9")
					sad_slider = gr.Slider(0, 1.0, value=0, step=0.05, label="😭 Sadness", info="Duration increased when beyond 0.2")
				with gr.Column():  # Input column
					happy_slider = gr.Slider(0, 1.0, value=0, step=0.05, label="😃 Happiness", info="Tread lightly beyond 0.7")
					surprise_slider = gr.Slider(0, 1.0, value=0, step=0.05, label="😮 Surprise", info="Can oversaturate Happiness")
			deepmoji_checkbox = gr.Checkbox(label="Use DeepMoji", info="Auto adjust emotional values", value=True)

	# Event handling using click
	btn = gr.Button("Generate")

	with gr.Row():  # Main row for inputs and language selection
		with gr.Column():  # Input column
			output_wav = gr.Audio(
				label="22kHz audio output",
				type="filepath",
				editable=False,
				autoplay=True
			)
		with gr.Column():  # Input column
			output_arpabet = gr.HTML(label="ARPAbet")

	btn.click(
		fn=predict,
		inputs=[
			input_textbox,
			voice_radio,
			language_radio,
			pacing_slider,
			pitch_slider,
			energy_slider,
			anger_slider,
			happy_slider,
			sad_slider,
			surprise_slider,
			deepmoji_checkbox
		],
		outputs=[
			output_wav,
			output_arpabet,
			anger_slider,
			happy_slider,
			sad_slider,
			surprise_slider,
			# xVAServer JSON
			gr.Textbox(visible=False)
		]
	)

	language_radio.change(
		set_default_text,
		inputs=[language_radio, deepmoji_checkbox],
		outputs=[input_textbox, deepmoji_checkbox]
	)

	deepmoji_checkbox.change(
		toggle_deepmoji,
		inputs=[
			deepmoji_checkbox,
			anger_slider,
			happy_slider,
			sad_slider,
			surprise_slider
		],
		outputs=[
			anger_slider,
			happy_slider,
			sad_slider,
			surprise_slider
		]
	)

	input_textbox.change(
		reset_em_sliders,
		inputs=[
			deepmoji_checkbox,
			anger_slider,
			happy_slider,
			sad_slider,
			surprise_slider
		],
		outputs=[
			anger_slider,
			happy_slider,
			sad_slider,
			surprise_slider
		]
	)

	voice_radio.change(
		reset_em_sliders,
		inputs=[
			deepmoji_checkbox,
			anger_slider,
			happy_slider,
			sad_slider,
			surprise_slider
		],
		outputs=[
			anger_slider,
			happy_slider,
			sad_slider,
			surprise_slider
		]
	)

if __name__ == "__main__":
	# Run the web server in a separate thread

	# print('Attempting to connect to local xVASynth server...')
	# try:
	# 	response = requests.get('http://0.0.0.0:8008')
	# 	response.raise_for_status()  # If the response contains an HTTP error status code, raise an exception
	# except requests.exceptions.RequestException as err:
	# 	print('Failed to connect to xVASynth!')
	# 	web_server_thread = threading.Thread(target=run_xvaserver)
	# 	print('Starting xVAServer thread')
	# 	web_server_thread.start()

	print('running Gradio interface')
	demo.launch()

	# Wait for the web server thread to finish (shouldn't be reached in normal execution)
	# web_server_thread.join()