feat: add ROCm + Coqui TTS with GPU support to Dockerfile

ai/Dockerfile

@@ -16,7 +16,8 @@ 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 && \
+        libportaudio2 \
+        hipcc espeak-ng && \
     rm -rf /var/lib/apt/lists/*
 
 # Création de l'utilisateur 'hermes' directement avec les bons accès
@@ -30,6 +31,75 @@ 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
 
@@ -41,6 +111,45 @@ COPY --chown=hermes:hermes . .
 RUN uv venv && \
     uv pip install --no-cache-dir sounddevice numpy faster-whisper
 
+# ---------- Patch tts_tool.py to add Coqui provider ----------
+RUN /opt/hermes/.venv/bin/python3 -c "
+tts_path = '/opt/hermes/.venv/lib/python3.13/site-packages/tools/tts_tool.py'
+with open(tts_path) as f:
+    code = f.read()
+coqui_block = '''
+        elif provider == \"coqui\":
+            logger.info(\"Generating speech with Coqui TTS (GPU, local)...\")
+            import subprocess
+            coqui_python = \"/opt/coqui-tts/bin/python3\"
+            coqui_script = \"/opt/coqui-tts/bin/coqui_synth.py\"
+            coqui_config = tts_config.get(\"coqui\", {})
+            model = coqui_config.get(\"model\", \"tts_models/en/vctk/vits\")
+            use_gpu = coqui_config.get(\"use_gpu\", True)
+            speaker = coqui_config.get(\"speaker\", \"\")
+            cmd = [
+                coqui_python, coqui_script,
+                \"--text\", text,
+                \"--out\", file_str,
+                \"--model\", model,
+            ]
+            if use_gpu:
+                cmd.append(\"--gpu\")
+            if speaker:
+                cmd.extend([\"--speaker\", speaker])
+            result = subprocess.run(cmd, capture_output=True, text=True, timeout=120)
+            if result.returncode != 0:
+                stderr = result.stderr.strip()
+                raise RuntimeError(f\"Coqui TTS failed: {stderr or 'unknown error'}\")
+            logger.info(\"Coqui TTS audio saved: %s\", file_str)
+'''
+code = code.replace(
+    '        else:\n            # Default: Edge TTS (free), with NeuTTS as local fallback',
+    coqui_block + '        else:\n            # Default: Edge TTS (free), with NeuTTS as local fallback'
+)
+with open(tts_path, 'w') as f:
+    f.write(code)
+"
+
 # ---------- Runtime ----------
 ENV HERMES_HOME=/opt/data
 ENV PATH="/opt/data/.local/bin:${PATH}"