update sofa_gui.py

Updates include:
- workaround to prevent trying to load an icon if running on linux
- add support for tgm_sofa_v004
- run in "forced" mode and not "match" mode (idk what they actually do but match is worse)

sofa_gui.py

@@ -1,12 +1,15 @@
-# GUI specific imports
+# general stuff
 import os, sys, re, glob
+# GUI stuff
 import customtkinter as ctk
 import tkinter as tk
-from ftfy import fix_text as fxy
+from ftfy import fix_text as fxy # unicode text all around fix
 import threading
+# function stuff
 import yaml
 import pathlib
 import warnings
+#import pykakasi # japanese language processing
 
 warnings.filterwarnings("ignore")
 sys.path.append('.')
@@ -28,10 +31,22 @@ def __init__(self, lang, wh_model):
 		self.tokenizer = get_tokenizer(multilingual=True)
 		self.number_tokens = [i for i in range(self.tokenizer.eot) if all(c in "0123456789" for c in self.tokenizer.decode([i]))]
 
-	def run_transcription(self, audio):
+	def conv_kana2roma(self, string):
+		# i have no idea how i want to implement this yet lol
+		#kks = pykakasi.kakasi():
+		#kks.setMode("J", "H")
+		#kks.setMode("K", "H")
+		#jpconv = kks.getConverter()
+		return 0
+
+	def run_transcription(self, audio, lang):
 		answer = self.model.transcribe(audio, suppress_tokens=[-1] + self.number_tokens)
+		if lang == "JP":
+			trns_str = fxy(self.conv_kana2roma(answer['text']))
+		else:
+			trns_str = fxy(answer['text'])
 		print(f"Wrote transcription for {audio} in corpus.")
-		return answer
+		return trns_str
 
 class App(ctk.CTk):
 	def __init__(self):
@@ -44,11 +59,11 @@ def whisper_function(self):
 			self.prog_bar.start()
 			trnsr = Transcriber(self.trans_lang_choice.get(), self.inf_wh_model.get())
 			try:
-				for file in glob.glob('corpus/**/*.wav'):
+				for file in glob.glob('corpus/**/*.wav', recursive=True):
 					answer = ''
 					out_name = file[:-4] + '.lab'
-					answer = trnsr.run_transcription(file)
-					output = answer['text'].lower()
+					trns_str = trnsr.run_transcription(file)
+					output = fxy(trns_str.lower())
 					final_op = re.sub(r"[.,!?]", "", output)
 					with open(out_name, 'w+', encoding='utf-8') as whis:
 						whis.write(final_op)
@@ -96,7 +111,7 @@ def infer_sofa(self, ckpt, dictionary, op_format):
 			# load model
 			torch.set_grad_enabled(False)
 			model = LitForcedAlignmentTask.load_from_checkpoint(ckpt)
-			model.set_inference_mode('match')
+			model.set_inference_mode('force')
 			trainer = pl.Trainer(logger=False)
 
 			# run predictions
@@ -194,7 +209,10 @@ def update_wh_model(self):
 		self.title(fxy(self._l[self.clang.get()]['app_ttl']))
 		self.geometry(f"{350}x{250}")
 		self.resizable(height=False, width=False)
-		self.wm_iconbitmap("assets/tgm.ico")
+
+		# including this to prevent error when running on Linux
+		if sys.platform == 'win32':
+			self.wm_iconbitmap("assets/tgm.ico")
 
 		# grid configs
 		self.grid_columnconfigure(0, weight=1)