use piper_phonemize as text tokenizer in ljspeech recipe

egs/ljspeech/TTS/local/prepare_token_file.py

@@ -17,7 +17,7 @@
 
 
 """
-This file reads the texts in given manifest and generates the file that maps tokens to IDs.
+This file generates the file that maps tokens to IDs.
 """
 
 import argparse
@@ -25,80 +25,47 @@ import logging
 from pathlib import Path
 from typing import Dict
 
-from lhotse import load_manifest
+from piper_phonemize import get_espeak_map
 
 
 def get_args():
     parser = argparse.ArgumentParser()
 
-    parser.add_argument(
-        "--manifest-file",
-        type=Path,
-        default=Path("data/spectrogram/ljspeech_cuts_train.jsonl.gz"),
-        help="Path to the manifest file",
-    )
-
     parser.add_argument(
         "--tokens",
         type=Path,
         default=Path("data/tokens.txt"),
-        help="Path to the tokens",
+        help="Path to the dict that maps the text tokens to IDs",
     )
 
     return parser.parse_args()
 
 
-def write_mapping(filename: str, sym2id: Dict[str, int]) -> None:
+def get_token2id(filename: Path) -> Dict[str, int]:
-    """Write a symbol to ID mapping to a file.
+    """Get a dict that maps token to IDs, and save it to the given filename."""
-
-    Note:
-      No need to implement `read_mapping` as it can be done
-      through :func:`k2.SymbolTable.from_file`.
-
-    Args:
-      filename:
-        Filename to save the mapping.
-      sym2id:
-        A dict mapping symbols to IDs.
-    Returns:
-      Return None.
-    """
-    with open(filename, "w", encoding="utf-8") as f:
-        for sym, i in sym2id.items():
-            f.write(f"{sym} {i}\n")
-
-
-def get_token2id(manifest_file: Path) -> Dict[str, int]:
-    """Return a dict that maps token to IDs."""
     extra_tokens = [
         "<blk>",  # 0 for blank
-        "<sos/eos>",  # 1 for sos and eos symbols.
+        "<sos>",  # 1 for sos
-        "<unk>",  # 2 for OOV
+        "<eos>",  # 2 for eos
+        "<unk>",  # 3 for OOV
     ]
-    all_tokens = set()
 
-    cut_set = load_manifest(manifest_file)
+    all_tokens = list(get_espeak_map().keys())
 
-    for cut in cut_set:
+    for t in extra_tokens:
-        # Each cut only contain one supervision
+        assert t not in all_tokens, t
-        assert len(cut.supervisions) == 1, len(cut.supervisions)
-        for t in cut.tokens:
-            all_tokens.add(t)
 
-    all_tokens = extra_tokens + list(all_tokens)
+    all_tokens = extra_tokens + all_tokens
 
-    token2id: Dict[str, int] = {token: i for i, token in enumerate(all_tokens)}
+    with open(filename, "w", encoding="utf-8") as f:
-    return token2id
+        for i, token in enumerate(all_tokens):
+            f.write(f"{token} {i}\n")
 
 
 if __name__ == "__main__":
     formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
-
     logging.basicConfig(format=formatter, level=logging.INFO)
 
     args = get_args()
-    manifest_file = Path(args.manifest_file)
     out_file = Path(args.tokens)
-
+    get_token2id(out_file)
-    token2id = get_token2id(manifest_file)
-    write_mapping(out_file, token2id)

egs/ljspeech/TTS/local/prepare_tokens_ljspeech.py

@@ -23,9 +23,9 @@ This file reads the texts in given manifest and save the new cuts with phoneme t
 import logging
 from pathlib import Path
 
-import g2p_en
 import tacotron_cleaner.cleaners
 from lhotse import CutSet, load_manifest
+from piper_phonemize import phonemize_espeak
 
 
 def prepare_tokens_ljspeech():
@@ -35,7 +35,6 @@ def prepare_tokens_ljspeech():
     partition = "all"
 
     cut_set = load_manifest(output_dir / f"{prefix}_cuts_{partition}.{suffix}")
-    g2p = g2p_en.G2p()
 
     new_cuts = []
     for cut in cut_set:
@@ -45,7 +44,11 @@ def prepare_tokens_ljspeech():
         # Text normalization
         text = tacotron_cleaner.cleaners.custom_english_cleaners(text)
         # Convert to phonemes
-        cut.tokens = g2p(text)
+        tokens_list = phonemize_espeak(text, "en-us")
+        tokens = []
+        for t in tokens_list:
+            tokens.extend(t)
+        cut.tokens = tokens
         new_cuts.append(cut)
 
     new_cut_set = CutSet.from_cuts(new_cuts)

egs/ljspeech/TTS/prepare.sh

@@ -30,7 +30,7 @@ if [ $stage -le -1 ] && [ $stop_stage -ge -1 ]; then
     cd vits/monotonic_align
     python setup.py build_ext --inplace
     cd ../../
-  else 
+  else
     log "monotonic_align lib already built"
   fi
 fi
@@ -80,6 +80,10 @@ fi
 
 if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
   log "Stage 3: Prepare phoneme tokens for LJSpeech"
+  # We assume you have installed piper_phonemize and espnet_tts_frontend.
+  # If not, please install them with:
+  #   - piper_phonemize: refer to https://github.com/rhasspy/piper-phonemize
+  #   - espnet_tts_frontend, `pip install espnet_tts_frontend`, refer to https://github.com/espnet/espnet_tts_frontend/
   if [ ! -e data/spectrogram/.ljspeech_with_token.done ]; then
     ./local/prepare_tokens_ljspeech.py
     mv data/spectrogram/ljspeech_cuts_with_tokens_all.jsonl.gz \
@@ -113,13 +117,11 @@ fi
 
 if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
   log "Stage 5: Generate token file"
-  # We assume you have installed g2p_en and espnet_tts_frontend.
+  # We assume you have installed piper_phonemize and espnet_tts_frontend.
   # If not, please install them with:
-  #   - g2p_en: `pip install g2p_en`, refer to https://github.com/Kyubyong/g2p
+  #   - piper_phonemize: refer to https://github.com/rhasspy/piper-phonemize
   #   - espnet_tts_frontend, `pip install espnet_tts_frontend`, refer to https://github.com/espnet/espnet_tts_frontend/
   if [ ! -e data/tokens.txt ]; then
-    ./local/prepare_token_file.py \
+    ./local/prepare_token_file.py --tokens data/tokens.txt
-      --manifest-file data/spectrogram/ljspeech_cuts_train.jsonl.gz \
-      --tokens data/tokens.txt
   fi
 fi

egs/ljspeech/TTS/vits/tokenizer.py

@@ -16,8 +16,8 @@
 
 from typing import Dict, List
 
-import g2p_en
 import tacotron_cleaner.cleaners
+from piper_phonemize import phonemize_espeak
 from utils import intersperse
 
 
@@ -41,18 +41,28 @@ class Tokenizer(object):
                 self.token2id[token] = id
 
         self.blank_id = self.token2id["<blk>"]
+        self.sos_id = self.token2id["<sos>"]
+        self.eos_id = self.token2id["<eos>"]
         self.oov_id = self.token2id["<unk>"]
         self.vocab_size = len(self.token2id)
 
-        self.g2p = g2p_en.G2p()
+    def texts_to_token_ids(
-
+        self,
-    def texts_to_token_ids(self, texts: List[str], intersperse_blank: bool = True):
+        texts: List[str],
+        intersperse_blank: bool = True,
+        add_sos: bool = False,
+        add_eos: bool = False,
+    ):
         """
         Args:
           texts:
             A list of transcripts.
           intersperse_blank:
             Whether to intersperse blanks in the token sequence.
+          add_sos:
+            Whether to add sos token at the start.
+          add_eos:
+            Whether to add eos token at the end.
 
         Returns:
           Return a list of token id list [utterance][token_id]
@@ -63,7 +73,11 @@ class Tokenizer(object):
             # Text normalization
             text = tacotron_cleaner.cleaners.custom_english_cleaners(text)
             # Convert to phonemes
-            tokens = self.g2p(text)
+            tokens_list = phonemize_espeak(text, "en-us")
+            tokens = []
+            for t in tokens_list:
+                tokens.extend(t)
+
             token_ids = []
             for t in tokens:
                 if t in self.token2id:
@@ -73,13 +87,21 @@ class Tokenizer(object):
 
             if intersperse_blank:
                 token_ids = intersperse(token_ids, self.blank_id)
+            if add_sos:
+                token_ids = [self.sos_id] + token_ids
+            if add_eos:
+                token_ids = token_ids + [self.eos_id]
 
             token_ids_list.append(token_ids)
 
         return token_ids_list
 
     def tokens_to_token_ids(
-        self, tokens_list: List[str], intersperse_blank: bool = True
+        self,
+        tokens_list: List[str],
+        intersperse_blank: bool = True,
+        add_sos: bool = False,
+        add_eos: bool = False,
     ):
         """
         Args:
@@ -87,6 +109,10 @@ class Tokenizer(object):
             A list of token list, each corresponding to one utterance.
           intersperse_blank:
             Whether to intersperse blanks in the token sequence.
+          add_sos:
+            Whether to add sos token at the start.
+          add_eos:
+            Whether to add eos token at the end.
 
         Returns:
           Return a list of token id list [utterance][token_id]
@@ -103,6 +129,10 @@ class Tokenizer(object):
 
             if intersperse_blank:
                 token_ids = intersperse(token_ids, self.blank_id)
+            if add_sos:
+                token_ids = [self.sos_id] + token_ids
+            if add_eos:
+                token_ids = token_ids + [self.eos_id]
 
             token_ids_list.append(token_ids)