github/ebook2audiobook
1
1
Fork 0
mirror of https://github.com/DrewThomasson/ebook2audiobook.git synced 2026-01-08 21:38:12 -05:00
Code Issues Packages Projects Releases Wiki Activity

v25.12.29

Browse Source
This commit is contained in:
ROBERT MCDOWELL
2025-12-28 21:02:08 -08:00
committed by GitHub
parent 426da60e1c d9292601a4
commit c3eb60c525
9 changed files with 53 additions and 122 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
VERSION.txt
View File

@@ -1 +1 @@
25.12.28
25.12.29

21
lib/classes/tts_engines/bark.py
View File

@@ -194,25 +194,12 @@ class Bark(TTSUtils, TTSRegistry, name='bark'):
**tts_dyn_params,
**fine_tuned_params
)
#audio_sentence = result.get('wav')
#if is_audio_data_valid(audio_sentence):
# audio_sentence = audio_sentence.tolist()
if is_audio_data_valid(audio_sentence):
if isinstance(audio_sentence, torch.Tensor):
audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
elif isinstance(audio_sentence, np.ndarray):
audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
elif isinstance(audio_sentence, (list, tuple)):
audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
else:
error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
print(error)
return False
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
sourceTensor = self._tensor_type(audio_sentence)
audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
if audio_tensor is not None and audio_tensor.numel() > 0:
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
self.audio_segments.append(audio_tensor)
if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

50
lib/classes/tts_engines/common/utils.py
View File

@@ -1,4 +1,5 @@
import os, threading, gc, torch, torchaudio, shutil, tempfile, regex as re, soundfile as sf, numpy as np
from lib.classes.tts_engines.common.audio import is_audio_data_valid
from typing import Any, Union, Dict
from huggingface_hub import hf_hub_download
@@ -210,36 +211,27 @@ class TTSUtils:
**fine_tuned_params,
)
audio_sentence = result.get('wav')
if isinstance(audio_sentence, torch.Tensor):
audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
elif isinstance(audio_sentence, np.ndarray):
audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
elif isinstance(audio_sentence, (list, tuple)):
audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
else:
error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
print(error)
return False
if audio_tensor is not None and audio_tensor.numel() > 0:
# CON is a reserved name on windows
lang_dir = 'con-' if self.session['language'] == 'con' else self.session['language']
new_voice_path = re.sub(r'([\\/])eng([\\/])', rf'\1{lang_dir}\2', voice_path)
proc_voice_path = new_voice_path.replace('.wav', '_temp.wav')
torchaudio.save(proc_voice_path, audio_tensor, default_engine_settings[xtts]['samplerate'], format='wav')
if normalize_audio(proc_voice_path, new_voice_path, default_audio_proc_samplerate, self.session['is_gui_process']):
del audio_sentence, audio_tensor
Path(proc_voice_path).unlink(missing_ok=True)
gc.collect()
self.engine = loaded_tts.get(self.tts_key, False)
if not self.engine:
self._load_engine()
return new_voice_path
if is_audio_data_valid(audio_sentence):
sourceTensor = self._tensor_type(audio_sentence)
audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
if audio_tensor is not None and audio_tensor.numel() > 0:
# CON is a reserved name on windows
lang_dir = 'con-' if self.session['language'] == 'con' else self.session['language']
new_voice_path = re.sub(r'([\\/])eng([\\/])', rf'\1{lang_dir}\2', voice_path)
proc_voice_path = new_voice_path.replace('.wav', '_temp.wav')
torchaudio.save(proc_voice_path, audio_tensor, default_engine_settings[xtts]['samplerate'], format='wav')
if normalize_audio(proc_voice_path, new_voice_path, default_audio_proc_samplerate, self.session['is_gui_process']):
del audio_sentence, sourceTensor, audio_tensor
Path(proc_voice_path).unlink(missing_ok=True)
gc.collect()
self.engine = loaded_tts.get(self.tts_key, False)
if not self.engine:
self._load_engine()
return new_voice_path
else:
error = 'normalize_audio() error:'
else:
error = 'normalize_audio() error:'
else:
error = f'No audio waveform found in _check_xtts_builtin_speakers() result: {result}'
error = f'No audio waveform found in _check_xtts_builtin_speakers() result: {result}'
else:
error = f"_check_xtts_builtin_speakers() error: {xtts} is False"
else:

18
lib/classes/tts_engines/fairseq.py
View File

@@ -166,21 +166,11 @@ class Fairseq(TTSUtils, TTSRegistry, name='fairseq'):
**speaker_argument
)
if is_audio_data_valid(audio_sentence):
if isinstance(audio_sentence, torch.Tensor):
audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
elif isinstance(audio_sentence, np.ndarray):
audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
elif isinstance(audio_sentence, (list, tuple)):
audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
else:
error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
print(error)
return False
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
sourceTensor = self._tensor_type(audio_sentence)
audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
if audio_tensor is not None and audio_tensor.numel() > 0:
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
self.audio_segments.append(audio_tensor)
if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

18
lib/classes/tts_engines/tacotron.py
View File

@@ -194,21 +194,11 @@ class Tacotron2(TTSUtils, TTSRegistry, name='tacotron'):
**speaker_argument
)
if is_audio_data_valid(audio_sentence):
if isinstance(audio_sentence, torch.Tensor):
audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
elif isinstance(audio_sentence, np.ndarray):
audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
elif isinstance(audio_sentence, (list, tuple)):
audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
else:
error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
print(error)
return False
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
sourceTensor = self._tensor_type(audio_sentence)
audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
if audio_tensor is not None and audio_tensor.numel() > 0:
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
self.audio_segments.append(audio_tensor)
if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

18
lib/classes/tts_engines/vits.py
View File

@@ -179,21 +179,11 @@ class Vits(TTSUtils, TTSRegistry, name='vits'):
**speaker_argument
)
if is_audio_data_valid(audio_sentence):
if isinstance(audio_sentence, torch.Tensor):
audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
elif isinstance(audio_sentence, np.ndarray):
audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
elif isinstance(audio_sentence, (list, tuple)):
audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
else:
error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
print(error)
return False
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
sourceTensor = self._tensor_type(audio_sentence)
audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
if audio_tensor is not None and audio_tensor.numel() > 0:
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
self.audio_segments.append(audio_tensor)
if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

18
lib/classes/tts_engines/xtts.py
View File

@@ -141,21 +141,11 @@ class XTTSv2(TTSUtils, TTSRegistry, name='xtts'):
)
audio_sentence = result.get('wav')
if is_audio_data_valid(audio_sentence):
if isinstance(audio_sentence, torch.Tensor):
audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
elif isinstance(audio_sentence, np.ndarray):
audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
elif isinstance(audio_sentence, (list, tuple)):
audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
else:
error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
print(error)
return False
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
sourceTensor = self._tensor_type(audio_sentence)
audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
if audio_tensor is not None and audio_tensor.numel() > 0:
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
self.audio_segments.append(audio_tensor)
if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

18
lib/classes/tts_engines/yourtts.py
View File

@@ -109,21 +109,11 @@ class YourTTS(TTSUtils, TTSRegistry, name='yourtts'):
**speaker_argument
)
if is_audio_data_valid(audio_sentence):
if isinstance(audio_sentence, torch.Tensor):
audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
elif isinstance(audio_sentence, np.ndarray):
audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
elif isinstance(audio_sentence, (list, tuple)):
audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
audio_tensor = audio_tensor.cpu()
else:
error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
print(error)
return False
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
sourceTensor = self._tensor_type(audio_sentence)
audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
if audio_tensor is not None and audio_tensor.numel() > 0:
if sentence[-1].isalnum() or sentence[-1] == '—':
audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
self.audio_segments.append(audio_tensor)
if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

12
lib/gradio.py
View File

@@ -1118,12 +1118,14 @@ def build_interface(args:dict)->gr.Blocks:
file_pattern = "*.wav"
eng_options = []
bark_options = []
builtin_dir = Path(os.path.join(voices_dir, lang_dir))
builtin_options = [
(os.path.splitext(f.name)[0], str(f))
for f in Path(os.path.join(voices_dir, lang_dir)).rglob(file_pattern)
(base, str(f))
for f in builtin_dir.rglob(file_pattern)
for base in [os.path.splitext(f.name)[0]]
]
builtin_names = {t[0]: None for t in builtin_options}
if session['language'] in default_engine_settings[TTS_ENGINES['XTTSv2']].get('languages', {}):
builtin_names = {t[0]: None for t in builtin_options}
eng_dir = Path(os.path.join(voices_dir, "eng"))
eng_options = [
(base, str(f))
@@ -1162,7 +1164,7 @@ def build_interface(args:dict)->gr.Blocks:
voice_options = [('Default', None)] + sorted(voice_options, key=lambda x: x[0].lower())
else:
voice_options = sorted(voice_options, key=lambda x: x[0].lower())
if session['voice'] is not None:
if session['voice'] is not None and isinstance(session.get('voice'), str):
if session['voice_dir'] not in session['voice']:
if not any(v[1] == session['voice'] for v in voice_options):
voice_path = Path(session['voice'])
@@ -1332,7 +1334,7 @@ def build_interface(args:dict)->gr.Blocks:
bark_visible = visible_gr_tab_bark_params
return (
gr.update(value=show_rating(session['tts_engine'])),
gr.update(visible=False),
gr.update(visible=visible_gr_tab_bark_params),
gr.update(visible=bark_visible),
gr.update(visible=False),
update_gr_fine_tuned_list(id),