NATURESPEAK-ML-UTTER/speak_metric_broadcast.py

import argparse, json, sys, time, random
import utterance.voice
import utterance.utils
import examine.metric

from osc4py3.as_eventloop import *
from osc4py3 import oscbuildparse

voices_broadcast = {}
conf = {}

def init_broadcast():

	global conf, voices_broadcast

	with open('voice.config.json') as f:
		conf = json.load(f)

	# configure voices and their channels
	for v in conf['voices']:
		voices_broadcast[v['name']] = v['osc_channel']['root'] + v['osc_channel']['utterance']

	print(voices_broadcast)		

	# Start the system.
	osc_startup()

	# Make client channels to send packets.
	osc_udp_client(conf['host'], conf['port'], "aaaa")

def broadcast_utterance(name, utterance):

	global conf, voices_broadcast

	sentences = utterance.split('\n')

	text = name + ":"

	for t in sentences:

		if ',' in t:
			phrases = t.split(',')
			for i in range(0, len(phrases)):
				p = phrases[i]				
				if p != "":
					text += " " + p
					if i != len(phrases) - 1 :
						text += ','
					msg = oscbuildparse.OSCMessage(voices_broadcast[name], None, [text])
					osc_send(msg, "aaaa")
					osc_process()
					time.sleep(2)
		else:
			if text == "":
				text += t
			else:
				text += "\n" + t
			msg = oscbuildparse.OSCMessage(voices_broadcast[name], None, [text])
			osc_send(msg, "aaaa")
			osc_process()
			time.sleep(2)



	msg = oscbuildparse.OSCMessage(conf['command_osc_channel'], None, ["clear"])
	osc_send(msg, "aaaa")
	osc_process()	


def format_str(text: str) -> str:
	t = utterance.utils.clean(text)
	return utterance.utils.format(t)	

				
def main() -> int:

	p = argparse.ArgumentParser()
	p.add_argument("-c", "--config", type=str, default="config.json", help="configuratin file")
	p.add_argument("-i", "--iterations", type=int, default=10, help="number of iterations")
	args = p.parse_args()

	print(args)

	with open(args.config) as f:
		conf = json.load(f)

	voices = []
	for v in conf['voices']:
		voice = utterance.voice.Voice(name=v["name"].upper(), model=v['model_dir'], tokenizer=v['tokeniser_file'], temp=float(v["temperature"]), lenght=7)
		voices.append(voice)

	init_broadcast()

	nbr_voices = len(voices)

	state = 'c'

	metric = examine.metric.Metric(model_input='data/models/doc2vec.model')

	s = set(range(0, nbr_voices - 1))

	# rindex = random.sample(s, 1)[0]
	rindex = 4

	v = voices[rindex]
	uv = v.utter_one()
	uv = format_str(uv)

	v_vec = metric.vector(uv)
	

	while 1:
		
		candidates = random.sample(s, 2)

		results = []
		for c in candidates:
			if c == rindex:
				continue
			vc = voices[c]
			vc_texts = vc.utter_n(n=25)
			for t in vc_texts:
				t = format_str(t)
				t_vec = metric.vector(t)
				d = examine.metric.cos_dist(v_vec, t_vec)
				results.append([d, t, c])


		results.sort(key=lambda t: t[0], reverse=True)

		# print('----------------------------')
		# print(v.name + ":")
		# print(uv)
		# print('----------------------------')

		broadcast_utterance(v.name, uv)

		# for r in results[:2]:
		# 	print('-->' + str(r[0]))
		# 	print(r[1])
		# 	print('+++++++++++++++++++++++++')


		# new round

		top = results[0]
		rindex = top[2]
		v = voices[rindex]
		uv = top[1]
		v_vec = metric.vector(top[1])


		# state = input("Continue? ")

	osc_terminate()


if __name__ == '__main__':
    sys.exit(main())