fix: ras Tokens need time&dur, not start&end

fs2/prediction_writing_callback.py

@@ -334,12 +334,11 @@ def __init__(
     def save_text(self, max_seconds, phones, words, language, filename):
         """Save the aligned text as a .readalong file"""
 
-        # Convert the (time, end, label) word tuples into RAS Tokens
         ras_tokens: list[Token] = []
-        for word in words:
+        for start, end, label in words:
             if ras_tokens:
-                ras_tokens.append(Token(" "))
-            ras_tokens.append(Token(word[2], word[0], word[1]))
+                ras_tokens.append(Token(text=" ", is_word=False))
+            ras_tokens.append(Token(text=label, time=start, dur=end - start))
 
         readalong = convert_to_readalong([ras_tokens], [language])
         with open(filename, "w", encoding="utf8") as f: