diff --git a/testing/fixtures/hermes-webui-0.52.181/static/panels.js b/testing/fixtures/hermes-webui-0.52.181/static/panels.js
new file mode 100644
index 00000000..856dd8b1
--- /dev/null
+++ b/testing/fixtures/hermes-webui-0.52.181/static/panels.js
@@ -0,0 +1,44 @@
+const _SETTINGS_SPEECH_STORAGE_KEYS={
+ tts_engine:'hermes-tts-engine',
+ tts_voice:'hermes-tts-voice',
+ tts_rate:'hermes-tts-rate',
+};
+let _settingsSpeechChangedKeys=new Set();
+
+function _speechPreferencesPayloadFromUi(){
+ const payload={};
+ const ttsVoiceSel=$('settingsTtsVoice');
+ if(ttsVoiceSel) _setOwnedSpeechPayload(payload,'tts_voice',ttsVoiceSel.value||'');
+ return payload;
+}
+
+function loadSettingsPanel(){
+ const ttsEngineSel=$('settingsTtsEngine');
+ if(ttsEngineSel){
+ ttsEngineSel.onchange=function(){
+ localStorage.setItem('hermes-tts-engine',this.value);
+ window._populateTtsVoices();
+ _schedulePreferencesAutosave();
+ };
+ }
+ // Populate voice selector based on engine
+ const ttsVoiceSel=$('settingsTtsVoice');
+ window._populateTtsVoices=function(){
+ if(!ttsVoiceSel) return;
+ const engine=localStorage.getItem('hermes-tts-engine')||'browser';
+ const current=String(_speechSetting('tts_voice','hermes-tts-voice','')||'');
+ _syncSpeechPreferenceCache('tts_voice',current);
+ if(engine==='edge'){
+ const edgeVoices=[
+ {value:'en-US-AriaNeural',label:'Aria (English, Female)'},
+ ];
+ ttsVoiceSel.innerHTML='';
+ edgeVoices.forEach(v=>ttsVoiceSel.appendChild(v));
+ }
+ };
+ if(ttsVoiceSel&&'speechSynthesis' in window){
+ window._populateTtsVoices();
+ ttsVoiceSel.onchange=function(){_markSpeechPreferenceChanged('tts_voice');localStorage.setItem('hermes-tts-voice',this.value);_schedulePreferencesAutosave();};
+ }
+ // TTS rate/pitch sliders
+}
diff --git a/testing/fixtures/hermes-webui-0.52.181/static/ui.js b/testing/fixtures/hermes-webui-0.52.181/static/ui.js
index e6546193..11ff370e 100644
--- a/testing/fixtures/hermes-webui-0.52.181/static/ui.js
+++ b/testing/fixtures/hermes-webui-0.52.181/static/ui.js
@@ -1,4 +1,15 @@
+function _buildBrowserUtterance(text, btn){
+ const utter=new SpeechSynthesisUtterance(text);
+ const savedVoice=localStorage.getItem('hermes-tts-voice');
+ const voices=speechSynthesis.getVoices();
+ if(savedVoice&&voices.length){
+ const match=voices.find(v=>v.name===savedVoice);
+ if(match) utter.voice=match;
+ }
+ return utter;
+}
function _playEdgeTtsChunked(text, btn){
+ const voice=localStorage.getItem('hermes-tts-voice')||'zh-CN-XiaoxiaoNeural';
return fetch('/api/tts',{body:JSON.stringify({text:chunk, voice:voice, rate:rate, pitch:pitch})});
}
function speakSelected(clean, btn, engine){
@@ -11,3 +22,14 @@ function speakAutomatically(clean, engine){
_playEdgeTtsChunked(clean, null);
}
}
+function registeredTts(engine, clean){
+ const _opts={
+ voice: localStorage.getItem('hermes-tts-voice')||'',
+ rate: parseFloat(localStorage.getItem('hermes-tts-rate')),
+ };
+ const autoOpts={
+ voice: localStorage.getItem('hermes-tts-voice')||'',
+ pitch: parseFloat(localStorage.getItem('hermes-tts-pitch')),
+ };
+ return [engine, clean, _opts, autoOpts];
+}
diff --git a/testing/tests/test_hermes_chat_quality.py b/testing/tests/test_hermes_chat_quality.py
index 4cf45ffb..87e08a53 100644
--- a/testing/tests/test_hermes_chat_quality.py
+++ b/testing/tests/test_hermes_chat_quality.py
@@ -398,13 +398,27 @@ def test_voice_models_are_baked_and_runtime_has_no_public_egress():
assert "HERMES_STT_CACHE=/opt/models/whisper" in stt_dockerfile
assert "ADD --checksum=sha256:4cabf7c3" in tts_dockerfile
assert "ADD --checksum=sha256:db42b97d" in tts_dockerfile
- assert tts_dockerfile.count("--chmod=0444") == 6
+ assert "ADD --checksum=sha256:b3a6e47b57b8c7fbe6a0ce2518161a50f59a9cdd8a50835c02cb02bdd6206c18" in tts_dockerfile
+ assert "ADD --checksum=sha256:95a23eb4d42909d38df73bb9ac7f45f597dbfcde2d1bf9526fdeaf5466977d77" in tts_dockerfile
+ assert "ADD --checksum=sha256:8ff38212d23da300bbe3705c645e6e5b9475f0bfde01558eb17813e22acaaaaa" in tts_dockerfile
+ assert "ADD --checksum=sha256:c2ec28bb38e2b59e93b959b3e40348c1afebbd272f30fed5d41205d08e98a9d7" in tts_dockerfile
+ assert "ADD --checksum=sha256:3ef40a71ea63852cd8ab7e6fa7d2ecdcfa67a0b47c9c48e3f10e02ee02083ea0" in tts_dockerfile
+ assert "ADD --checksum=sha256:1afc81f703c0e4cb3b4d7c0dca096b8b54a98806807f0170cf5eb5557723c12d" in tts_dockerfile
+ assert tts_dockerfile.count("--chmod=0444") == 12
+ assert "/opt/models/piper/en_US-amy-medium.onnx" in tts_dockerfile
+ assert "/opt/models/piper/ru_RU-irina-medium.onnx" in tts_dockerfile
+ assert "/opt/models/piper/es_MX-claude-high.onnx" in tts_dockerfile
assert "chmod 0555 /opt/models /opt/models/piper" in tts_dockerfile
assert "HERMES_TTS_CACHE=/opt/models/piper" in tts_dockerfile
+ assert "HERMES_TTS_VOICE=en_US-amy-medium" in tts_dockerfile
tts_server = (ROOT / "dockerfiles" / "hermes-jetson-tts-server.py").read_text()
assert "download_voice" not in tts_server
assert "baked Piper voice is missing" in tts_server
- assert "session_options.intra_op_num_threads = ONNX_THREADS" in tts_server
+ assert "session_options.intra_op_num_threads = threads" in tts_server
+ assert 'LANGUAGE_VOICE_MAP = {' in tts_server
+ assert '"en": "en_US-amy-medium"' in tts_server
+ assert '"ru": "ru_RU-irina-medium"' in tts_server
+ assert '"es": "es_MX-claude-high"' in tts_server
policies = _documents(HERMES / "networkpolicy.yaml")
voice_policy = next(
diff --git a/testing/tests/test_hermes_tts_language_routing.py b/testing/tests/test_hermes_tts_language_routing.py
new file mode 100644
index 00000000..5a53f808
--- /dev/null
+++ b/testing/tests/test_hermes_tts_language_routing.py
@@ -0,0 +1,220 @@
+"""Language allow-list contracts for the private Hermes chat TTS voice policy."""
+
+from __future__ import annotations
+
+import importlib.util
+import io
+import json
+import sys
+from types import SimpleNamespace
+
+import pytest
+
+from testing.tests.test_hermes_chat_support import ROOT
+
+AMY = "en_US-amy-medium"
+IRINA = "ru_RU-irina-medium"
+CLAUDE = "es_MX-claude-high"
+
+
+def _load_tts_server(monkeypatch):
+ server_path = ROOT / "dockerfiles" / "hermes-jetson-tts-server.py"
+ spec = importlib.util.spec_from_file_location("hermes_jetson_tts_server", server_path)
+ assert spec and spec.loader
+ module = importlib.util.module_from_spec(spec)
+
+ class _FakeSessionOptions:
+ def __init__(self) -> None:
+ self.intra_op_num_threads = None
+ self.inter_op_num_threads = None
+
+ fake_onnxruntime = SimpleNamespace(
+ SessionOptions=_FakeSessionOptions,
+ InferenceSession=lambda *a, **k: SimpleNamespace(),
+ )
+ fake_piper = SimpleNamespace(
+ PiperConfig=SimpleNamespace(from_dict=lambda d: d),
+ PiperVoice=lambda **kwargs: SimpleNamespace(**kwargs),
+ SynthesisConfig=lambda **kwargs: SimpleNamespace(**kwargs),
+ )
+ monkeypatch.setitem(sys.modules, "onnxruntime", fake_onnxruntime)
+ monkeypatch.setitem(sys.modules, "piper", fake_piper)
+ spec.loader.exec_module(module)
+ return module
+
+
+@pytest.fixture
+def tts(monkeypatch):
+ return _load_tts_server(monkeypatch)
+
+
+@pytest.mark.parametrize(
+ "language,expected",
+ [
+ ("en", AMY),
+ ("en-US", AMY),
+ ("en_US", AMY),
+ ("EN", AMY),
+ ("En-Us", AMY),
+ ("ru", IRINA),
+ ("ru-RU", IRINA),
+ ("ru_RU", IRINA),
+ ("RU", IRINA),
+ ("es", CLAUDE),
+ ("es-MX", CLAUDE),
+ ("es_MX", CLAUDE),
+ ("es-ES", CLAUDE),
+ ("es_ES", CLAUDE),
+ ("ES", CLAUDE),
+ ],
+)
+def test_allow_listed_languages_resolve_to_the_approved_voice(tts, language, expected):
+ assert tts.resolve_voice_name(language) == expected
+
+
+@pytest.mark.parametrize(
+ "language",
+ [
+ None,
+ "",
+ " ",
+ "fr",
+ "fr-FR",
+ "de-DE",
+ "xx",
+ "en-GB",
+ "es-AR",
+ "english",
+ 123,
+ 1.5,
+ True,
+ [],
+ {},
+ {"lang": "ru"},
+ "../../etc/passwd",
+ "en_US-amy-medium/../../ru_RU-irina-medium",
+ "\x00ru",
+ "ru\x00",
+ ],
+)
+def test_unknown_missing_or_malformed_language_falls_back_to_amy(tts, language):
+ assert tts.resolve_voice_name(language) == AMY
+
+
+def test_default_voice_name_matches_the_dockerfile_env_default(tts):
+ assert tts.DEFAULT_VOICE_NAME == AMY
+
+
+def test_resolved_voice_is_always_one_of_the_three_baked_names(tts):
+ assert frozenset({AMY, IRINA, CLAUDE}) == tts.BAKED_VOICE_NAMES
+ fuzz_inputs = [
+ "en", "ru", "es", "unknown", "", None, 42, "../../../etc/shadow",
+ "en_US-amy-medium\x00; rm -rf /", "RU-ru", "Es-Es", "en-us-extra",
+ ]
+ for value in fuzz_inputs:
+ assert tts.resolve_voice_name(value) in tts.BAKED_VOICE_NAMES
+
+
+def test_client_voice_field_cannot_override_the_language_policy(tts):
+ """The POST handler must select the voice from "language" only.
+
+ A malicious or stale "voice" field in a hostile/legacy request must never
+ change which baked model answers the request.
+ """
+ calls: list[str] = []
+
+ class _RecordingVoice:
+ def __init__(self, name: str) -> None:
+ self.name = name
+
+ def synthesize_wav(self, text, wav_file, syn_config) -> None:
+ calls.append(self.name)
+ wav_file.setnchannels(1)
+ wav_file.setsampwidth(2)
+ wav_file.setframerate(16_000)
+ wav_file.writeframes(b"\x00\x00")
+
+ class _RecordingHandler(tts.SpeechHandler):
+ def __init__(self, payload):
+ request = json.dumps(payload).encode("utf-8")
+ self.path = "/v1/audio/speech"
+ self.headers = {"Content-Length": str(len(request))}
+ self.rfile = io.BytesIO(request)
+ self.wfile = io.BytesIO()
+ self.status = None
+ self.response_headers = {}
+ self.server = SimpleNamespace(
+ voices={
+ AMY: _RecordingVoice(AMY),
+ IRINA: _RecordingVoice(IRINA),
+ CLAUDE: _RecordingVoice(CLAUDE),
+ },
+ default_voice_name=AMY,
+ )
+
+ def send_response(self, status, message=None):
+ self.status = status
+
+ def send_header(self, name, value):
+ self.response_headers[name] = value
+
+ def end_headers(self):
+ return None
+
+ # A payload that supplies an attacker/legacy "voice" value but no
+ # language must resolve to the safe default, never the "voice" value.
+ handler = _RecordingHandler({"input": "hi", "voice": IRINA})
+ handler.do_POST()
+ assert handler.status == 200
+ assert handler.response_headers["X-TTS-Voice"] == AMY
+
+ # A payload supplying both must still be governed by "language" alone.
+ handler = _RecordingHandler({"input": "hi", "voice": CLAUDE, "language": "ru"})
+ handler.do_POST()
+ assert handler.status == 200
+ assert handler.response_headers["X-TTS-Voice"] == IRINA
+
+ assert calls == [AMY, IRINA]
+
+
+def test_no_client_string_reaches_a_filesystem_path(tts):
+ """resolve_voice_name must only ever return a fixed, baked literal.
+
+ This is the property that keeps a client from ever causing the server to
+ build a Path out of attacker-controlled text: the return value is always
+ a member of the fixed allow-list, regardless of input shape.
+ """
+ hostile_inputs = [
+ "../../../../etc/passwd",
+ "/etc/passwd",
+ "en_US-amy-medium/../../../etc/passwd",
+ "ru_RU-irina-medium\x00.onnx",
+ "es_MX-claude-high; cat /etc/shadow",
+ "\n\ren",
+ "en" + "/" * 200,
+ " ",
+ ]
+ for value in hostile_inputs:
+ result = tts.resolve_voice_name(value)
+ assert result in tts.BAKED_VOICE_NAMES
+ assert "/" not in result
+ assert ".." not in result
+ assert "\x00" not in result
+
+
+def test_normalize_language_rejects_non_string_input(tts):
+ assert tts.normalize_language(None) is None
+ assert tts.normalize_language(123) is None
+ assert tts.normalize_language([]) is None
+ assert tts.normalize_language("") is None
+ assert tts.normalize_language(" ") is None
+ assert tts.normalize_language("En_US") == "en-us"
+
+
+def test_default_voice_name_is_one_of_the_baked_voices(tts):
+ assert tts.DEFAULT_VOICE_NAME in tts.BAKED_VOICE_NAMES
+
+
+def test_load_voices_fails_closed_when_a_baked_model_is_missing(tts, tmp_path):
+ with pytest.raises(RuntimeError, match="baked Piper voice is missing"):
+ tts.load_voices(tmp_path, threads=1)
diff --git a/testing/tests/test_hermes_voice_instrument.py b/testing/tests/test_hermes_voice_instrument.py
index eada0199..b458a5f8 100644
--- a/testing/tests/test_hermes_voice_instrument.py
+++ b/testing/tests/test_hermes_voice_instrument.py
@@ -57,6 +57,55 @@ def test_real_upstream_fixture_receives_visual_instrument_contract(tmp_path: Pat
assert '