preparations for training an IPA-based finetune

This commit is contained in:
mrq 2023-03-16 04:25:33 +00:00
parent 7b80f7a42f
commit ee8270bdfb
5 changed files with 172 additions and 23 deletions

121
models/tokenizers/ipa.json Executable file
View File

@ -0,0 +1,121 @@
{
"version": "1.0",
"truncation": null,
"padding": null,
"added_tokens":
[
{
"id": 0,
"special": true,
"content": "[STOP]",
"single_word": false,
"lstrip": false,
"rstrip": false,
"normalized": false
},
{
"id": 1,
"special": true,
"content": "[UNK]",
"single_word": false,
"lstrip": false,
"rstrip": false,
"normalized": false
},
{
"id": 2,
"special": true,
"content": "[SPACE]",
"single_word": false,
"lstrip": false,
"rstrip": false,
"normalized": false
}
],
"normalizer": null,
"pre_tokenizer": null,
"post_processor": null,
"decoder": null,
"model":
{
"type": "BPE",
"dropout": null,
"unk_token": "[UNK]",
"continuing_subword_prefix": null,
"end_of_word_suffix": null,
"fuse_unk": false,
"vocab":
{
"[STOP]": 0,
"[UNK]": 1,
"[SPACE]": 2,
"!": 3,
"'": 4,
"(": 5,
")": 6,
",": 7,
"-": 8,
".": 9,
"/": 10,
":": 11,
";": 12,
"?": 13,
"a": 14,
"aɪ": 15,
"aʊ": 16,
"b": 17,
"d": 18,
"d͡": 19,
"d͡ʒ": 20,
"e": 21,
"eɪ": 22,
"f": 23,
"h": 24,
"i": 25,
"j": 26,
"k": 27,
"l": 28,
"m": 29,
"n": 30,
"o": 31,
"oʊ": 32,
"p": 33,
"s": 34,
"t": 35,
"t͡": 36,
"t͡ʃ": 37,
"u": 38,
"v": 39,
"w": 40,
"z": 41,
"|": 42,
"æ": 43,
"ð": 44,
"ŋ": 45,
"ɑ": 46,
"ɔ": 47,
"ɔɪ": 48,
"ə": 49,
"ɚ": 50,
"ɛ": 51,
"ɡ": 52,
"ɪ": 53,
"ɹ": 54,
"ʃ": 55,
"ʊ": 56,
"ʌ": 57,
"ʒ": 58,
"θ": 59
},
"merges":
[
"a ɪ",
"a ʊ",
"d͡ ʒ",
"e ɪ",
"o ʊ",
"t͡ ʃ",
ɪ"
]
}
}

@ -1 +1 @@
Subproject commit b253da6e353f0170c3eb60fe299c41d2fa21db50
Subproject commit 730a04708d2cb29f526c3397894950a2733e6e29

@ -1 +1 @@
Subproject commit 42cb1f36741aa3a24e7aab03e73b51becd182fa7
Subproject commit 99618694db4cd7b77e68b62753bb8e2418ac0d55

View File

@ -20,8 +20,7 @@ import subprocess
import psutil
import yaml
import hashlib
import io
import gzip
import string
import tqdm
import torch
@ -40,6 +39,13 @@ from tortoise.utils.text import split_and_recombine_text
from tortoise.utils.device import get_device_name, set_device_name, get_device_count, get_device_vram, get_device_batch_size, do_gc
from whisper.normalizers.english import EnglishTextNormalizer
from whisper.normalizers.basic import BasicTextNormalizer
from whisper.tokenizer import LANGUAGES
try:
from phonemizer import phonemize as phonemizer
except Exception as e:
pass
MODELS['dvae.pth'] = "https://huggingface.co/jbetker/tortoise-tts-v2/resolve/3704aea61678e7e468a06d8eea121dba368a798e/.models/dvae.pth"
@ -64,7 +70,7 @@ VALLE_ENABLED = False
try:
from vall_e.emb.qnt import encode as quantize
from vall_e.emb.g2p import encode as phonemize
# from vall_e.emb.g2p import encode as phonemize
VALLE_ENABLED = True
except Exception as e:
@ -1157,7 +1163,6 @@ def transcribe_dataset( voice, language=None, skip_existings=False, progress=Non
if whisper_model is None:
load_whisper_model(language=language)
results = {}
files = sorted( get_voices(load_latents=False)[voice] )
@ -1175,14 +1180,15 @@ def transcribe_dataset( voice, language=None, skip_existings=False, progress=Non
if basename in results and skip_existings:
print(f"Skipping already parsed file: {basename}")
else:
results[basename] = whisper_transcribe(file, language=language)
result = whisper_transcribe(file, language=language)
results[basename] = result
waveform, sample_rate = torchaudio.load(file)
# resample to the input rate, since it'll get resampled for training anyways
# this should also "help" increase throughput a bit when filling the dataloaders
waveform, sample_rate = resample(waveform, sample_rate, tts.input_sample_rate if tts is not None else 22050)
torchaudio.save(f"{indir}/audio/{basename}", waveform, sample_rate)
torchaudio.save(f"{indir}/audio/{basename}", waveform, sample_rate, encoding="PCM_S", bits_per_sample=16)
with open(infile, 'w', encoding="utf-8") as f:
f.write(json.dumps(results, indent='\t'))
@ -1248,18 +1254,28 @@ def slice_dataset( voice, trim_silence=True, start_offset=0, end_offset=0, resul
messages.append(message)
continue
sliced, _ = resample( sliced, sample_rate, 22050 )
torchaudio.save(f"{indir}/audio/{file}", sliced, 22050)
torchaudio.save(f"{indir}/audio/{file}", sliced, 22050, encoding="PCM_S", bits_per_sample=16)
segments +=1
messages.append(f"Sliced segments: {files} => {segments}.")
return "\n".join(messages)
def prepare_dataset( voice, use_segments, text_length, audio_length, normalize=True ):
"""
def phonemizer( text, language="eng" ):
transducer = make_g2p(language, f'{language}-ipa')
phones = transducer(text).output_string
ignored = [" "] + [ p for p in string.punctuation ]
return ["_" if p in ignored else p for p in phones]
"""
def prepare_dataset( voice, use_segments=False, text_length=0, audio_length=0, normalize=True ):
indir = f'./training/{voice}/'
infile = f'{indir}/whisper.json'
messages = []
phonemize = phonemize=args.tokenizer_json[-8:] == "ipa.json"
if not os.path.exists(infile):
raise Exception(f"Missing dataset: {infile}")
@ -1272,12 +1288,19 @@ def prepare_dataset( voice, use_segments, text_length, audio_length, normalize=T
'supervisions': [],
}
normalizer = EnglishTextNormalizer() if normalize else None
errored = 0
for filename in results:
result = results[filename]
use_segment = use_segments
result = results[filename]
language = LANGUAGES[result['language']] if result['language'] in LANGUAGES else None
if language == "english":
language = "en-us"
normalizer = None
if normalize:
normalizer = EnglishTextNormalizer() if language.lower()[:2] == "en" else BasicTextNormalizer()
# check if unsegmented text exceeds 200 characters
if not use_segment:
@ -1325,7 +1348,14 @@ def prepare_dataset( voice, use_segments, text_length, audio_length, normalize=T
continue
text = segment['text'].strip()
normalized_text = normalizer(text) if normalize and result['language'] == "en" else text
normalized_text = normalizer(text) if normalize else None
try:
phonemes = phonemizer( text, language=language, preserve_punctuation=True, strip=True ) if phonemize else None
except Exception as e:
pass
if phonemize and phonemes:
text = phonemes
if len(text) > 200:
message = f"Text length too long (200 < {len(text)}), skipping... {file}"
@ -1351,11 +1381,7 @@ def prepare_dataset( voice, use_segments, text_length, audio_length, normalize=T
if not culled and audio_length > 0:
culled = duration < audio_length
# for when i add in a little treat ;), as it requires normalized text
if normalize and len(normalized_text) < 200:
line = f'audio/{file}|{text}|{normalized_text}'
else:
line = f'audio/{file}|{text}'
line = f'audio/{file}|{text}'
lines['training' if not culled else 'validation'].append(line)
@ -1365,7 +1391,7 @@ def prepare_dataset( voice, use_segments, text_length, audio_length, normalize=T
os.makedirs(f'{indir}/valle/', exist_ok=True)
from vall_e.emb.qnt import encode as quantize
from vall_e.emb.g2p import encode as phonemize
# from vall_e.emb.g2p import encode as phonemize
if waveform.shape[0] == 2:
waveform = waveform[:1]
@ -1373,8 +1399,8 @@ def prepare_dataset( voice, use_segments, text_length, audio_length, normalize=T
quantized = quantize( waveform, sample_rate ).cpu()
torch.save(quantized, f'{indir}/valle/{file.replace(".wav",".qnt.pt")}')
phonemes = phonemize(normalized_text)
open(f'{indir}/valle/{file.replace(".wav",".phn.txt")}', 'w', encoding='utf-8').write(" ".join(phonemes))
# phonemes = phonemizer(normalized_text)
open(f'{indir}/valle/{file.replace(".wav",".phn.txt")}', 'w', encoding='utf-8').write(" ".join(text))
training_joined = "\n".join(lines['training'])
validation_joined = "\n".join(lines['validation'])
@ -1536,8 +1562,10 @@ def save_training_settings( **kwargs ):
if settings['save_rate'] < 1:
settings['save_rate'] = 1
"""
if settings['validation_rate'] < 1:
settings['validation_rate'] = 1
"""
settings['validation_batch_size'] = int(settings['batch_size'] / settings['gradient_accumulation_size'])
@ -1554,7 +1582,6 @@ def save_training_settings( **kwargs ):
settings['validation_enabled'] = False
messages.append("Validation batch size == 0, disabling validation...")
else:
settings['validation_enabled'] = True
with open(settings['validation_path'], 'r', encoding="utf-8") as f:
validation_lines = len(f.readlines())

View File

@ -443,7 +443,7 @@ def setup_gradio():
DATASET_SETTINGS['validation_text_length'] = gr.Number(label="Validation Text Length Threshold", value=12, precision=0)
DATASET_SETTINGS['validation_audio_length'] = gr.Number(label="Validation Audio Length Threshold", value=1 )
with gr.Row():
DATASET_SETTINGS['skip'] = gr.Checkbox(label="Skip Already Transcribed", value=False)
DATASET_SETTINGS['skip'] = gr.Checkbox(label="Skip Existing", value=False)
DATASET_SETTINGS['slice'] = gr.Checkbox(label="Slice Segments", value=False)
DATASET_SETTINGS['trim_silence'] = gr.Checkbox(label="Trim Silence", value=False)
with gr.Row():
@ -496,6 +496,7 @@ def setup_gradio():
with gr.Row():
TRAINING_SETTINGS["half_p"] = gr.Checkbox(label="Half Precision", value=args.training_default_halfp)
TRAINING_SETTINGS["bitsandbytes"] = gr.Checkbox(label="BitsAndBytes", value=args.training_default_bnb)
TRAINING_SETTINGS["validation_enabled"] = gr.Checkbox(label="Validation Enabled", value=False)
with gr.Row():
TRAINING_SETTINGS["workers"] = gr.Number(label="Worker Processes", value=2, precision=0)