Merge pull request 'feat: add voiceclone audio API and bridge bootstrap' (#19) from feat/tts-generate-20260813125855 into main

app.py

@@ -1,13 +1,13 @@
 """Homelab Butler v2.1 – Unified API proxy for Pfannkuchen homelab.
 Reads service config from butler.yaml, credentials from Vaultwarden cache with flat-file fallback."""
 
-import os, json, asyncio, logging, time, base64, re, subprocess, ipaddress
+import os, json, asyncio, logging, time, base64, re, subprocess, ipaddress, secrets
 from datetime import datetime, timezone
 import httpx, yaml
 from typing import Literal
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
 from fastapi import FastAPI, Request, HTTPException, Depends, Query
-from fastapi.responses import JSONResponse, RedirectResponse
+from fastapi.responses import JSONResponse, RedirectResponse, Response
 from contextlib import asynccontextmanager
 
 log = logging.getLogger("butler")
@@ -273,6 +273,7 @@ async def root():
             "inventory_add": "POST /inventory/host {name, ip, group?}",
             "ansible_run": "POST /ansible/run {hostname}",
             "tts_speak": "POST /tts/speak {text, target: speaker|telegram}",
+            "tts_generate": "POST /tts/generate {text, voice?, language?} - return cloned WAV audio",
             "tts_voices": "GET /tts/voices",
             "tts_health": "GET /tts/health",
             "status": "GET /status - health of all backends",
@@ -1834,9 +1835,98 @@ class TTSRequest(BaseModel):
     voice: str = "deep_thought.mp3"
     language: str = "de"
 
+
+class TTSGenerateRequest(BaseModel):
+    text: str = Field(min_length=1, max_length=2000)
+    voice: str = Field(default="deep_thought.mp3", pattern=r"^[A-Za-z0-9_.-]+$")
+    language: str = Field(default="de", pattern=r"^[A-Za-z]{2,8}(?:-[A-Za-z0-9]{2,8})?$")
+
+
+class TTSBridgeDeployRequest(BaseModel):
+    rotate_client_token: bool = False
+
+
 SPEAKER_URL = TTS_CFG.get("speaker_url", "http://10.10.1.166:10800") if TTS_CFG else "http://10.10.1.166:10800"
 CHATTERBOX_URL = TTS_CFG.get("chatterbox_url", "http://10.2.1.104:8004/tts") if TTS_CFG else "http://10.2.1.104:8004/tts"
 
+
+def _chatterbox_payload(text: str, voice: str, language: str) -> dict:
+    return {
+        "text": text,
+        "voice_mode": "clone",
+        "reference_audio_filename": voice,
+        "output_format": "wav",
+        "language": language,
+        "exaggeration": 0.3,
+        "cfg_weight": 0.7,
+        "temperature": 0.6,
+    }
+
+
+@app.post("/tts/generate", response_class=Response)
+async def tts_generate(req: TTSGenerateRequest, _=Depends(_verify)):
+    """Generate cloned speech and return the WAV bytes to the authenticated caller."""
+    async with httpx.AsyncClient(verify=False, timeout=180) as client:
+        try:
+            result = await client.post(CHATTERBOX_URL, json=_chatterbox_payload(req.text, req.voice, req.language))
+        except httpx.RequestError:
+            _audit("/tts/generate", "POST", 502, "chatterbox request failed")
+            raise HTTPException(status_code=502, detail="Chatterbox is unavailable")
+
+    if result.status_code != 200:
+        _audit("/tts/generate", "POST", 502, f"chatterbox_http={result.status_code}")
+        raise HTTPException(status_code=502, detail="Chatterbox generation failed")
+    if not result.content.startswith(b"RIFF"):
+        _audit("/tts/generate", "POST", 502, "invalid audio response")
+        raise HTTPException(status_code=502, detail="Chatterbox returned invalid audio")
+
+    _audit("/tts/generate", "POST", 200, f"voice={req.voice} chars={len(req.text)}")
+    return Response(
+        content=result.content,
+        media_type="audio/wav",
+        headers={
+            "Content-Disposition": 'inline; filename="voiceclone.wav"',
+            "Cache-Control": "no-store",
+            "X-Content-Type-Options": "nosniff",
+        },
+    )
+
+
+@app.post("/tts/bridge/deploy")
+async def tts_bridge_deploy(req: TTSBridgeDeployRequest, _=Depends(_verify)):
+    """Install host-local bridge secrets on automation1 without exposing them."""
+    client_token = _vault_cache.get("tts_bridge_client_token", "").strip()
+    if req.rotate_client_token or not client_token:
+        client_token = secrets.token_urlsafe(32)
+    if not BUTLER_TOKEN:
+        raise HTTPException(status_code=500, detail="Butler token is not configured")
+
+    files = {
+        "/app-config/tts-bridge/butler-token": BUTLER_TOKEN,
+        "/app-config/tts-bridge/client-token": client_token,
+    }
+    encoded = base64.b64encode(json.dumps(files).encode()).decode()
+    script = (
+        "import base64,json,os,pathlib;"
+        f"files=json.loads(base64.b64decode('{encoded}'));"
+        "pathlib.Path('/app-config/tts-bridge').mkdir(parents=True,exist_ok=True);"
+        "[(pathlib.Path(p).write_text(v),os.chmod(p,0o600)) for p,v in files.items()]"
+    )
+    rc, _out, err = _ssh("sascha@10.5.85.5", f"python3 -c {__import__('shlex').quote(script)}", timeout=30)
+    if rc != 0:
+        _audit("/tts/bridge/deploy", "POST", 500, "secret installation failed")
+        raise HTTPException(status_code=500, detail="Could not install bridge secrets")
+
+    _audit("/tts/bridge/deploy", "POST", 200, f"rotated={req.rotate_client_token or not _vault_cache.get('tts_bridge_client_token')}")
+    return {
+        "status": "ready",
+        "host": "automation1",
+        "listen": "0.0.0.0:8099",
+        "client_token": client_token,
+        "rotated": req.rotate_client_token or not _vault_cache.get("tts_bridge_client_token"),
+    }
+
+
 @app.post("/tts/speak")
 async def tts_speak(req: TTSRequest, _=Depends(_verify)):
     if req.target == "speaker":

tests/test_app.py

@@ -45,6 +45,110 @@ def test_health_exposes_current_version():
     assert response.json()["version"] == app.VERSION == "2.3.5"
 
 
+def test_tts_generate_returns_cloned_wav(monkeypatch):
+    captured = {}
+
+    class FakeResponse:
+        status_code = 200
+        content = b"RIFF" + b"test-wave"
+
+    class FakeClient:
+        def __init__(self, **_kwargs):
+            pass
+
+        async def __aenter__(self):
+            return self
+
+        async def __aexit__(self, *_args):
+            return False
+
+        async def post(self, url, json):
+            captured["url"] = url
+            captured["json"] = json
+            return FakeResponse()
+
+    monkeypatch.setattr(app.httpx, "AsyncClient", FakeClient)
+    with TestClient(app.app) as client:
+        response = client.post(
+            "/tts/generate",
+            headers={"Authorization": "Bearer test-token"},
+            json={"text": "Hallo Sascha", "voice": "deep_thought.mp3", "language": "de"},
+        )
+    assert response.status_code == 200
+    assert response.headers["content-type"].startswith("audio/wav")
+    assert response.content.startswith(b"RIFF")
+    assert captured["json"]["voice_mode"] == "clone"
+    assert captured["json"]["reference_audio_filename"] == "deep_thought.mp3"
+
+
+def test_tts_generate_validates_text_and_voice_before_backend(monkeypatch):
+    monkeypatch.setattr(
+        app.httpx,
+        "AsyncClient",
+        lambda **_kwargs: (_ for _ in ()).throw(AssertionError("backend must not be called")),
+    )
+    with TestClient(app.app) as client:
+        empty = client.post(
+            "/tts/generate",
+            headers={"Authorization": "Bearer test-token"},
+            json={"text": ""},
+        )
+        traversal = client.post(
+            "/tts/generate",
+            headers={"Authorization": "Bearer test-token"},
+            json={"text": "Hallo", "voice": "../secret.wav"},
+        )
+    assert empty.status_code == 422
+    assert traversal.status_code == 422
+
+
+def test_tts_generate_rejects_non_wav_backend_response(monkeypatch):
+    class FakeResponse:
+        status_code = 200
+        content = b"not audio"
+
+    class FakeClient:
+        def __init__(self, **_kwargs):
+            pass
+
+        async def __aenter__(self):
+            return self
+
+        async def __aexit__(self, *_args):
+            return False
+
+        async def post(self, _url, json):
+            return FakeResponse()
+
+    monkeypatch.setattr(app.httpx, "AsyncClient", FakeClient)
+    with TestClient(app.app) as client:
+        response = client.post(
+            "/tts/generate",
+            headers={"Authorization": "Bearer test-token"},
+            json={"text": "Hallo"},
+        )
+    assert response.status_code == 502
+    assert "invalid audio" in response.text
+
+
+def test_tts_bridge_deploy_installs_secrets_without_logging_them(monkeypatch):
+    calls = []
+    monkeypatch.setattr(app, "BUTLER_TOKEN", "butler-secret")
+    monkeypatch.setattr(app, "_vault_cache", {})
+    monkeypatch.setattr(app, "_ssh", lambda host, command, timeout=30: (calls.append((host, command, timeout)) or (0, "", "")))
+    with TestClient(app.app) as client:
+        response = client.post(
+            "/tts/bridge/deploy",
+            headers={"Authorization": "Bearer butler-secret"},
+            json={"rotate_client_token": True},
+        )
+    assert response.status_code == 200
+    assert response.json()["listen"] == "0.0.0.0:8099"
+    assert len(response.json()["client_token"]) >= 32
+    assert calls[0][0] == "sascha@10.5.85.5"
+    assert "butler-secret" not in calls[0][1]
+
+
 def test_sysctl_audit_reads_fixed_keys_from_inventory_host(monkeypatch):
     payload = {
         "live": {"net.ipv4.tcp_congestion_control": "bbr"},