From c312343585daf7da2ddd275d37da446419cb4367 Mon Sep 17 00:00:00 2001
From: Jindrich Matousek <jindrich.matousek@gmail.com>
Date: Wed, 6 Sep 2023 17:05:47 +0200
Subject: [PATCH] Language of each item (sample/utterance) is set to dataset
 language only when not defined at the sample/utterance level Speaker name is
 prepended by dataset name in case of multispeaker datasets Refactor "artic"
 formatter

---
 TTS/tts/datasets/__init__.py   |  8 ++++++--
 TTS/tts/datasets/formatters.py | 13 ++++++++-----
 2 files changed, 14 insertions(+), 7 deletions(-)

diff --git a/TTS/tts/datasets/__init__.py b/TTS/tts/datasets/__init__.py
index 19213856..02434917 100644
--- a/TTS/tts/datasets/__init__.py
+++ b/TTS/tts/datasets/__init__.py
@@ -58,8 +58,12 @@ def split_dataset(items, eval_split_max_size=None, eval_split_size=0.01):
 
 def add_extra_keys(metadata, language, dataset_name):
     for item in metadata:
-        # add language name
-        item["language"] = language
+        # JMa: Add language name only if not defined at the sample level. Could  be good for multi-language datasets.
+        if not item["language"]:
+            item["language"] = language
+        # JMa: Prepend dataset name to speaker name. Could be good for multispeaker datasets.
+        if item["speaker_name"] != dataset_name and not item["speaker_name"].startswith(dataset_name+"_"):
+            item["speaker_name"] = f'{dataset_name}_{item["speaker_name"]}'
         # add unique audio name
         relfilepath = os.path.splitext(os.path.relpath(item["audio_file"], item["root_path"]))[0]
         audio_unique_name = f"{dataset_name}#{relfilepath}"
diff --git a/TTS/tts/datasets/formatters.py b/TTS/tts/datasets/formatters.py
index 2ea75ec3..113fc8d1 100644
--- a/TTS/tts/datasets/formatters.py
+++ b/TTS/tts/datasets/formatters.py
@@ -652,11 +652,14 @@ def artic(root_path, meta_file, **kwargs):  # pylint: disable=unused-argument
     items = []
     # Speaker name is the name of the directory with the data (last part of `root_path`)
     speaker_name = os.path.basename(os.path.normpath(root_path))
-    # Speaker name can consists of language code (eg. cs-CZ) and gender (m/f) separated by dots
-    # Example: AndJa.cs-CZ.m
-    parts = speaker_name.split(".")
-    lang = parts[1] if len(parts) == 3 and "-" in parts[1] else None
-    print(f" > ARTIC dataset: voice {parts[0]}, language {lang}")
+    # Speaker name can consists of language code (eg. cs-CZ or en) and gender (m/f) separated by dots
+    # Example: AndJa.cs-CZ.m, LJS.en.f
+    try:
+        voice, lang, sex = speaker_name.split(".")
+    except ValueError:
+        voice = speaker_name
+        lang, sex = None, None
+    print(f" > ARTIC dataset: voice={voice}, sex={sex}, language={lang}")
     with open(txt_file, "r", encoding="utf-8") as ttf:
         for line in ttf:
             # Check the number of standard separators