fix: replace Coqui/ROCm with Piper TTS (simpler, local, CPU)
This commit is contained in:
@@ -16,8 +16,7 @@ RUN apt-get update && \
|
||||
texlive-latex-base texlive-latex-extra texlive-fonts-recommended texlive-xetex texlive-science \
|
||||
qemu-user-static binfmt-support qemu-user-binfmt \
|
||||
emacs-nox \
|
||||
libportaudio2 \
|
||||
hipcc espeak-ng && \
|
||||
libportaudio2 && \
|
||||
rm -rf /var/lib/apt/lists/*
|
||||
|
||||
# Création de l'utilisateur 'hermes' directement avec les bons accès
|
||||
@@ -31,74 +30,6 @@ WORKDIR /opt/hermes
|
||||
# On donne la propriété du dossier de travail à l'utilisateur hermes
|
||||
RUN chown hermes:hermes /opt/hermes
|
||||
|
||||
# ---------- Coqui TTS venv (Python 3.11 + PyTorch ROCm) ----------
|
||||
# Install Python 3.11 via uv for Coqui compatibility
|
||||
RUN uv python install 3.11
|
||||
|
||||
# Create the coqui venv and install PyTorch ROCm + TTS
|
||||
RUN uv venv --python 3.11 /opt/coqui-tts
|
||||
RUN uv pip install --python /opt/coqui-tts/bin/python3 --no-cache-dir \
|
||||
torch==2.3.1+rocm5.7 \
|
||||
torchaudio==2.3.1+rocm5.7 \
|
||||
--index-url https://download.pytorch.org/whl/rocm5.7
|
||||
RUN uv pip install --python /opt/coqui-tts/bin/python3 --no-cache-dir TTS setuptools
|
||||
|
||||
# Fix executable stack on bundled torch AMD libraries (required for ROCm)
|
||||
RUN /opt/coqui-tts/bin/python3 -c "
|
||||
import struct, os, glob
|
||||
torch_lib = '/opt/coqui-tts/lib/python3.11/site-packages/torch/lib'
|
||||
for so in glob.glob(os.path.join(torch_lib, '*.so*')):
|
||||
try:
|
||||
with open(so, 'r+b') as f:
|
||||
if f.read(4) != b'\x7fELF': continue
|
||||
f.seek(0)
|
||||
h = f.read(64)
|
||||
e_phoff = struct.unpack_from('<Q', h, 0x20)[0]
|
||||
e_phentsize = struct.unpack_from('<H', h, 0x36)[0]
|
||||
e_phnum = struct.unpack_from('<H', h, 0x38)[0]
|
||||
for i in range(e_phnum):
|
||||
f.seek(e_phoff + i * e_phentsize)
|
||||
ph = f.read(e_phentsize)
|
||||
if struct.unpack_from('<I', ph, 0)[0] == 0x6474e551:
|
||||
flags = struct.unpack_from('<I', ph, 4)[0]
|
||||
if flags & 1:
|
||||
f.seek(e_phoff + i * e_phentsize + 4)
|
||||
f.write(struct.pack('<I', flags - 1))
|
||||
break
|
||||
except: pass
|
||||
"
|
||||
|
||||
# Créer le wrapper Coqui synth
|
||||
RUN /opt/coqui-tts/bin/python3 -c "
|
||||
script = '''#!/usr/bin/env python3
|
||||
import argparse, sys
|
||||
def main():
|
||||
parser = argparse.ArgumentParser()
|
||||
parser.add_argument('--text', required=True)
|
||||
parser.add_argument('--out', required=True)
|
||||
parser.add_argument('--model', default='tts_models/en/vctk/vits')
|
||||
parser.add_argument('--speaker', default='')
|
||||
parser.add_argument('--gpu', action='store_true', default=True)
|
||||
args = parser.parse_args()
|
||||
try:
|
||||
from TTS.api import TTS
|
||||
tts = TTS(model_name=args.model, progress_bar=False, gpu=args.gpu)
|
||||
kwargs = {'text': args.text, 'file_path': args.out}
|
||||
if args.speaker and hasattr(tts, 'speakers') and tts.speakers:
|
||||
kwargs['speaker'] = args.speaker
|
||||
tts.tts_to_file(**kwargs)
|
||||
print('OK:' + args.out)
|
||||
except Exception as e:
|
||||
print('ERROR:' + str(e), file=sys.stderr)
|
||||
sys.exit(1)
|
||||
if __name__ == '__main__':
|
||||
main()
|
||||
'''
|
||||
with open('/opt/coqui-tts/bin/coqui_synth.py', 'w') as f:
|
||||
f.write(script)
|
||||
import os; os.chmod('/opt/coqui-tts/bin/coqui_synth.py', 0o755)
|
||||
"
|
||||
|
||||
# ---------- Hermes venv ----------
|
||||
# Passer immédiatement sous l'utilisateur hermes pour tout le reste du build
|
||||
USER hermes
|
||||
|
||||
Reference in New Issue
Block a user