Merge pull request 'feat(hermes): Piper TTS (local US male, no cloud)' (#17) from feat/voice-support-v2 into master

Reviewed-on: #17

ai/Dockerfile

@@ -1,71 +1,73 @@
-FROM ghcr.io/astral-sh/uv:0.11.6-python3.13-trixie@sha256:b3c543b6c4f23a5f2df22866bd7857e5d304b67a564f4feab6ac22044dde719b AS uv_source
+# 1. On récupère la version la plus récente d'UV
-FROM tianon/gosu:1.19-trixie@sha256:3b176695959c71e123eb390d427efc665eeb561b1540e82679c15e992006b8b9 AS gosu_source
+FROM ghcr.io/astral-sh/uv:latest AS uv_source
-FROM debian:13.4
 
-# Disable Python stdout buffering to ensure logs are printed immediately
+# 2. Image officielle Hermes Agent de NousResearch
-ENV PYTHONUNBUFFERED=1
+# Contient déjà: Python, Node.js, npm, Playwright/Chromium, venv, tts_tool.py, etc.
+FROM nousresearch/hermes-agent:latest
 
-# Store Playwright browsers outside the volume mount so the build-time
+# ---------- System dependencies ----------
-# install survives the /opt/data volume overlay at runtime.
+# The official hermes-agent image already has: git, curl, ffmpeg, python3,
-ENV PLAYWRIGHT_BROWSERS_PATH=/opt/hermes/.playwright
+# gcc, build-essential, openssh-client, procps, tini, ripgrep, docker-cli,
-
+# libportaudio2, ca-certificates, etc.
-# Install system dependencies in one layer, clear APT cache
+#
-# tini reaps orphaned zombie processes (MCP stdio subprocesses, git, bun, etc.)
+# These extras we need to add back:
-# that would otherwise accumulate when hermes runs as PID 1. See #15012.
+#   - poppler-utils, imagemagick  (PDF/image processing)
+#   - texlive-*                   (LaTeX typesetting for reports)
+#   - qemu-user-static, binfmt-support (QEMU cross-compilation)
+#   - emacs-nox                   (text editing in container)
+USER root
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
-        build-essential nodejs npm python3 ripgrep ffmpeg gcc python3-dev libffi-dev procps git openssh-client docker-cli tini \
+        libportaudio2 \
-        curl poppler-utils imagemagick \
+        ca-certificates \
-        chromium xvfb fonts-noto-color-emoji fonts-unifont fonts-liberation fonts-ipafont-gothic fonts-wqy-zenhei fonts-tlwg-loma-otf fonts-freefont-ttf \
+        poppler-utils \
-        libasound2t64 libatk-bridge2.0-0t64 libatk1.0-0t64 libatspi2.0-0t64 libcairo2 libcups2t64 libdbus-1-3 libdrm2 libgbm1 libglib2.0-0t64 libnspr4 libnss3 libpango-1.0-0 libx11-6 libxcb1 libxcomposite1 libxdamage1 libxext6 libxfixes3 libxkbcommon0 libxrandr2 \
+        imagemagick \
-        texlive-latex-base texlive-latex-extra texlive-fonts-recommended texlive-xetex texlive-science \
+        texlive-latex-base \
-        qemu-user-static binfmt-support qemu-user-binfmt \
+        texlive-latex-extra \
-        emacs-nox \
+        texlive-fonts-recommended \
-        libportaudio2 && \
+        texlive-xetex \
+        texlive-science \
+        qemu-user-static \
+        binfmt-support \
+        emacs-nox && \
     rm -rf /var/lib/apt/lists/*
 
-# Non-root user for runtime; UID can be overridden via HERMES_UID at runtime
+# ---------- UV (hyperfast pip alternative) ----------
-RUN useradd -u 10000 -m -d /opt/data hermes
+COPY --chmod=0755 --from=uv_source /uv /usr/local/bin/
-
-COPY --chmod=0755 --from=gosu_source /gosu /usr/local/bin/
-COPY --chmod=0755 --from=uv_source /usr/local/bin/uv /usr/local/bin/uvx /usr/local/bin/
 
 WORKDIR /opt/hermes
 
-# ---------- Layer-cached dependency install ----------
+# ---------- Piper TTS dans le venv existant ----------
-# Copy only package manifests first so npm install + Playwright are cached
+# Le venv de l'image de base est root-owned, on doit installer en root aussi
-# unless the lockfiles themselves change.
+RUN . /opt/hermes/.venv/bin/activate && \
-COPY package.json package-lock.json ./
+    uv pip install --no-cache-dir piper-tts sounddevice numpy
-COPY web/package.json web/package-lock.json web/
 
-RUN npm install --prefer-offline --no-audit && \
+# ---------- Télécharger la voix Piper Ryan (high quality) ----------
-    npx playwright install --with-deps chromium --only-shell && \
+RUN mkdir -p /opt/hermes/.venv/share/piper/voices && \
-    (cd web && npm install --prefer-offline --no-audit) && \
+    /opt/hermes/.venv/bin/python3 /dev/stdin << 'PYEOF'
-    npm cache clean --force
+import urllib.request
+base = '/opt/hermes/.venv/share/piper/voices'
+url = 'https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/ryan/high/en_US-ryan-high.onnx'
+urllib.request.urlretrieve(url, base + '/en_US-ryan-high.onnx')
+urllib.request.urlretrieve(url + '.json', base + '/en_US-ryan-high.onnx.json')
+PYEOF
 
-# ---------- Source code ----------
+# ---------- Patch tts_tool.py: remplacer Edge TTS par Piper ----------
-# .dockerignore excludes node_modules, so the installs above survive.
+# Edge TTS appelle les serveurs Microsoft — on ne veut jamais ça.
-COPY --chown=hermes:hermes . .
+# Piper roule localement sur CPU, aucun cloud, aucune donnée qui sort.
-
+COPY patch_tts_tool.py /tmp/patch_tts_tool.py
-# Build web dashboard (Vite outputs to hermes_cli/web_dist/)
+RUN /opt/hermes/.venv/bin/python3 /tmp/patch_tts_tool.py && rm /tmp/patch_tts_tool.py
-RUN cd web && npm run build
-
-# ---------- Permissions ----------
-# Make install dir world-readable so any HERMES_UID can read it at runtime.
-# The venv needs to be traversable too.
-USER root
-RUN chmod -R a+rX /opt/hermes
-# Start as root so the entrypoint can usermod/groupmod + gosu.
-# If HERMES_UID is unset, the entrypoint drops to the default hermes user (10000).
-
-# ---------- Python virtualenv ----------
-RUN uv venv && \
-    uv pip install --no-cache-dir -e ".[all]" && \
-    uv pip install --no-cache-dir sounddevice numpy faster-whisper
 
 # ---------- Runtime ----------
-ENV HERMES_WEB_DIST=/opt/hermes/hermes_cli/web_dist
+# Retour à l'utilisateur non-root pour la sécurité
+USER hermes
+
 ENV HERMES_HOME=/opt/data
 ENV PATH="/opt/data/.local/bin:${PATH}"
+
 VOLUME [ "/opt/data" ]
-ENTRYPOINT [ "/usr/bin/tini", "-g", "--", "/opt/hermes/docker/entrypoint.sh" ]
+
+# Script de réparation des permissions + patch TTS au démarrage
+COPY --chmod=0755 fix-permissions.sh /opt/hermes/fix-permissions.sh
+
+ENTRYPOINT [ "/usr/bin/tini", "-g", "--", "/opt/hermes/fix-permissions.sh" ]

ai/compose.yml

@@ -39,6 +39,12 @@ services:
       - API_SERVER_KEY=hermes_local_key
       - GATEWAY_ALLOW_ALL_USERS=true
       - OPENROUTER_API_KEY=${OPENROUTER_API_KEY}
+      # ROCm for GPU-accelerated faster-whisper STT
+      - HSA_OVERRIDE_GFX_VERSION=9.0.6
+      - HCC_AMDGPU_TARGET=gfx906
+      - HIP_VISIBLE_DEVICES=0,1
+      - ROCR_VISIBLE_DEVICES=0,1
+      - HSA_ENABLE_SDMA=0
     volumes:
       - /mnt/HoardingCow_docker_data/Hermes/data:/opt/data
     devices:

ai/fix-permissions.sh

@@ -0,0 +1,38 @@
+#!/bin/bash
+# Startup permission fix + TTS patch.
+# Runs as root before the entrypoint drops to the hermes user.
+set -e
+
+HERMES_HOME="${HERMES_HOME:-/opt/data}"
+
+# Fix ownership on critical writable directories
+chown -R hermes:hermes \
+  "$HERMES_HOME/sessions" \
+  "$HERMES_HOME/checkpoints" \
+  "$HERMES_HOME/skills" \
+  "$HERMES_HOME/memories" \
+  "$HERMES_HOME/workspace" \
+  "$HERMES_HOME/pastes" \
+  "$HERMES_HOME/logs" \
+  "$HERMES_HOME/cron" \
+  "$HERMES_HOME/plans" \
+  "$HERMES_HOME/hooks" \
+  "$HERMES_HOME/cache" \
+  2>/dev/null || true
+
+# Fix data volume root ownership
+if [ "$(stat -c %u "$HERMES_HOME" 2>/dev/null)" != "$(id -u hermes)" ]; then
+  chown hermes:hermes "$HERMES_HOME" 2>/dev/null || true
+fi
+
+# ---------- Patch tts_tool.py: replace Edge TTS with Piper ----------
+# Fallback runtime patch in case the volume's site-packages differ from the image.
+# Idempotent: if already patched, the script does nothing.
+PATCH_SCRIPT="/opt/hermes/patch_tts_tool.py"
+if [ -f "$PATCH_SCRIPT" ]; then
+  echo "Applying TTS patch (Piper only, no Edge fallback)..."
+  /opt/hermes/.venv/bin/python3 "$PATCH_SCRIPT" 2>&1 || true
+fi
+
+# Chain to the official Hermes entrypoint
+exec /opt/hermes/docker/entrypoint.sh "$@"

ai/patch_tts_tool.py

@@ -0,0 +1,181 @@
+#!/usr/bin/env python3
+"""Patch Hermes TTS tool: add Piper TTS provider, remove Edge TTS as default.
+
+Patches ALL copies of tts_tool.py found (venv site-packages + /opt/hermes/tools/).
+
+Searches multiple paths for tts_tool.py so it works both at build time
+(in the image venv) and at runtime (on the mounted data volume).
+
+Idempotent: if already patched, does nothing.
+"""
+
+import sys
+import os
+
+# ---------------------------------------------------------------------------
+# Search for all copies of tts_tool.py
+# ---------------------------------------------------------------------------
+CANDIDATE_PATHS = [
+    "/opt/hermes/.venv/lib/python3.13/site-packages/tools/tts_tool.py",
+    "/opt/hermes/tools/tts_tool.py",
+]
+
+found_paths = []
+
+for p in CANDIDATE_PATHS:
+    if os.path.exists(p):
+        found_paths.append(p)
+        print(f"Found tts_tool.py at: {p}")
+
+# Also try to find via Python import
+import subprocess
+try:
+    result = subprocess.run(
+        [sys.executable, "-c", "import tools.tts_tool; print(tools.tts_tool.__file__)"],
+        capture_output=True, text=True, timeout=5
+    )
+    if result.returncode == 0:
+        p = result.stdout.strip()
+        if os.path.exists(p) and p not in found_paths:
+            found_paths.append(p)
+            print(f"Found tts_tool.py via import at: {p}")
+except Exception:
+    pass
+
+if not found_paths:
+    print("WARNING: tts_tool.py not found anywhere. Patching deferred to runtime.")
+    print(f"Searched: {CANDIDATE_PATHS}")
+    sys.exit(0)
+
+# ---------------------------------------------------------------------------
+# Old else block: the Edge TTS default fallback to replace
+# ---------------------------------------------------------------------------
+old_else = '''        else:
+            # Default: Edge TTS (free), with NeuTTS as local fallback
+            edge_available = True
+            try:
+                _import_edge_tts()
+            except ImportError:
+                edge_available = False
+
+            if edge_available:
+                logger.info("Generating speech with Edge TTS...")
+                try:
+                    import concurrent.futures
+                    with concurrent.futures.ThreadPoolExecutor(max_workers=1) as pool:
+                        pool.submit(
+                            lambda: asyncio.run(_generate_edge_tts(text, file_str, tts_config))
+                        ).result(timeout=60)
+                except RuntimeError:
+                    asyncio.run(_generate_edge_tts(text, file_str, tts_config))
+            elif _check_neutts_available():
+                logger.info("Edge TTS not available, falling back to NeuTTS (local)...")
+                provider = "neutts"
+                _generate_neutts(text, file_str, tts_config)
+            else:
+                return json.dumps({
+                    "success": False,
+                    "error": "No TTS provider available. Install edge-tts (pip install edge-tts) "
+                             "or set up NeuTTS for local synthesis."
+                }, ensure_ascii=False)'''
+
+# ---------------------------------------------------------------------------
+# New block: elif provider == "piper" + else: fallback with Piper only
+# ---------------------------------------------------------------------------
+new_block = '''        elif provider == "piper":
+            # Piper TTS (local, CPU, no cloud, no Microsoft)
+            piper_binary = "/opt/hermes/.venv/bin/piper"
+            piper_config = tts_config.get("piper", {})
+            voice = piper_config.get("voice", "en_US-lessac-medium")
+            model_dir = piper_config.get("model_dir", "/opt/hermes/.venv/share/piper/voices")
+            model_path = os.path.join(model_dir, f"{voice}.onnx")
+            if not os.path.exists(model_path):
+                return json.dumps({
+                    "success": False,
+                    "error": "Piper TTS voice model not found. "
+                             "Install Piper TTS and download a voice model."
+                }, ensure_ascii=False)
+            logger.info("Generating speech with Piper TTS (local, CPU)...")
+            import subprocess as _sp
+            cmd = [piper_binary, "--model", model_path, "--output-raw"]
+            try:
+                proc = _sp.Popen(cmd, stdin=_sp.PIPE, stdout=_sp.PIPE, stderr=_sp.PIPE)
+                raw_audio, stderr = proc.communicate(input=text.encode(), timeout=60)
+                if proc.returncode != 0:
+                    raise RuntimeError(f"Piper TTS failed: {stderr.decode()[:200]}")
+                ffmpeg_cmd = ["ffmpeg", "-f", "s16le", "-ar", "22050", "-ac", "1", "-i", "-", "-y", file_str]
+                _sp.run(ffmpeg_cmd, input=raw_audio, capture_output=True, timeout=30)
+            except Exception as e:
+                return json.dumps({
+                    "success": False,
+                    "error": f"Piper TTS failed: {e}"
+                }, ensure_ascii=False)
+
+        else:
+            # Default: Piper TTS (local, CPU, no cloud, no Microsoft)
+            piper_binary = "/opt/hermes/.venv/bin/piper"
+            piper_config = tts_config.get("piper", {})
+            voice = piper_config.get("voice", "en_US-lessac-medium")
+            model_dir = piper_config.get("model_dir", "/opt/hermes/.venv/share/piper/voices")
+            model_path = os.path.join(model_dir, f"{voice}.onnx")
+            if os.path.exists(model_path) and os.path.exists(piper_binary):
+                logger.info("Generating speech with Piper TTS (local, CPU)...")
+                import subprocess as _sp
+                cmd = [piper_binary, "--model", model_path, "--output-raw"]
+                try:
+                    proc = _sp.Popen(cmd, stdin=_sp.PIPE, stdout=_sp.PIPE, stderr=_sp.PIPE)
+                    raw_audio, stderr = proc.communicate(input=text.encode(), timeout=60)
+                    if proc.returncode != 0:
+                        raise RuntimeError(stderr.decode()[:200])
+                    ffmpeg_cmd = ["ffmpeg", "-f", "s16le", "-ar", "22050", "-ac", "1", "-i", "-", "-y", file_str]
+                    _sp.run(ffmpeg_cmd, input=raw_audio, capture_output=True, timeout=30)
+                except Exception:
+                    pass
+            else:
+                return json.dumps({
+                    "success": False,
+                    "error": "Piper TTS not available. Install piper-tts and download a voice model."
+                }, ensure_ascii=False)'''
+
+# ---------------------------------------------------------------------------
+# Apply the patch to all copies found
+# ---------------------------------------------------------------------------
+patched_any = False
+
+for tts_path in found_paths:
+    with open(tts_path) as f:
+        code = f.read()
+
+    if 'provider == "piper"' in code:
+        print(f"ALREADY PATCHED: {tts_path}")
+        continue
+
+    if old_else in code:
+        code = code.replace(old_else, new_block, 1)
+        with open(tts_path, 'w') as f:
+            f.write(code)
+        print(f"PATCHED: {tts_path}")
+        patched_any = True
+    else:
+        print(f"SKIP {tts_path}: Edge fallback pattern not found")
+        import re
+        for m in re.finditer(r'        else:\n            # Default:', code):
+            start = max(0, m.start() - 100)
+            end = min(len(code), m.end() + 300)
+            print(f"  Found 'else:/# Default:' at position {m.start()}:")
+            print(f"  {code[start:end]}")
+            print("  ---")
+        # Don't exit with error — if one copy isn't patchable, try the others
+
+if not patched_any:
+    all_patched = all(
+        'provider == "piper"' in open(p).read()
+        for p in found_paths
+    )
+    if all_patched:
+        print("All copies already patched.")
+        sys.exit(0)
+    print("WARNING: Could not patch any copy of tts_tool.py")
+    sys.exit(1)
+
+print("tts_tool.py patched successfully across all copies.")