From d31f27119ab634261d97f111d14762b1ec8b9187 Mon Sep 17 00:00:00 2001
From: mrq <mrq@ecker.tech>
Date: Sat, 21 Sep 2024 12:29:28 -0500
Subject: [PATCH] regex replace out the (lang) markers in espeak, updated
 tokenizer vocab as lazily as possible to not have unk tokens

---
 data/tokenizer.json | 13 ++++++++++++-
 vall_e/data.py      |  6 ++++--
 2 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/data/tokenizer.json b/data/tokenizer.json
index 66bd5e0..892771a 100644
--- a/data/tokenizer.json
+++ b/data/tokenizer.json
@@ -116,12 +116,16 @@
       "<mask>": 3,
       " ": 4,
       
-      "ᵝ": 0,
+      "ᵝ": 4,
 
       "!": 5,
       "\"": 6,
       "(": 7,
+      "{": 4,
+      "[": 4,
       ")": 8,
+      "}": 8,
+      "]": 8,
       ",": 9,
       "-": 10,
       ".": 11,
@@ -131,6 +135,7 @@
       "?": 15,
       "a": 16,
       "ä": 16,
+      "ɒ": 16,
       "b": 17,
       "c": 18,
       "d": 19,
@@ -145,6 +150,7 @@
       "m": 27,
       "n": 28,
       "ɴ": 28,
+      "ɲ": 28,
       "o": 29,
       "̞": 29,
       "p": 30,
@@ -152,9 +158,14 @@
       "q": 31,
       "r": 32,
       "ɽ": 32,
+      "ʁ": 32,
       "s": 33,
       "t": 34,
       "u": 35,
+      "ø": 35,
+      "œ": 35,
+      "y": 35,
+      "ɣ": 35,
       "ũ": 35,
       "v": 36,
       "w": 37,
diff --git a/vall_e/data.py b/vall_e/data.py
index fb2b0f2..6bea50d 100755
--- a/vall_e/data.py
+++ b/vall_e/data.py
@@ -3,6 +3,7 @@
 import copy
 import h5py
 import json
+import re
 import logging
 import numpy as np
 import os
@@ -1370,8 +1371,9 @@ def process_artifact_metadata( artifact ):
 	# to-do: regex replace /([a-z]{2})/ to ""
 	if "phonemes" in metadata:
 		metadata["phonemes"] = metadata["phonemes"].replace("(en)", "")
-	if "phonemes" in metadata and "language" in metadata:
-		metadata["phonemes"] = metadata["phonemes"].replace(f"({metadata['language']})", "")
+		if "language" in metadata:
+			metadata["phonemes"] = metadata["phonemes"].replace(f"({metadata['language']})", "")
+		metadata["phonemes"] = re.sub(r'\([a-z]{2}\)', "", metadata["phonemes"])
 
 	return metadata