verbalcoding 0.2.11 → 0.2.13

package/app-node/tts_settings.test.mjs

@@ -121,6 +121,210 @@ test('buildTtsSettings normalizes Supertonic local backend settings', () => {
   assert.equal(settings.supertonic.interOpThreads, '1');
 });
 
+test('buildTtsSettings normalizes OmniVoice local backend settings', () => {
+  const root = '/project';
+  const settings = buildTtsSettings({
+    TTS_BACKEND: 'omnivoice',
+    OMNIVOICE_PYTHON: './.venv-omnivoice/bin/python',
+    OMNIVOICE_MODEL: 'k2-fsa/OmniVoice',
+    OMNIVOICE_DEVICE: 'mps',
+    OMNIVOICE_DTYPE: 'float16',
+    OMNIVOICE_REF_AUDIO: './voice-samples/me.wav',
+    OMNIVOICE_REF_TEXT: '테스트 기준 음성입니다.',
+    OMNIVOICE_LANGUAGE: 'ko',
+    OMNIVOICE_SPEAKER: 'warm korean male voice',
+    OMNIVOICE_TIMEOUT_MS: '180000',
+    OMNIVOICE_PROGRESS: '1',
+  }, root);
+
+  assert.equal(settings.backend, 'omnivoice');
+  assert.equal(settings.omnivoice.python, path.join(root, '.venv-omnivoice', 'bin', 'python'));
+  assert.equal(settings.omnivoice.model, 'k2-fsa/OmniVoice');
+  assert.equal(settings.omnivoice.device, 'mps');
+  assert.equal(settings.omnivoice.dtype, 'float16');
+  assert.equal(settings.omnivoice.refAudio, path.join(root, 'voice-samples', 'me.wav'));
+  assert.equal(settings.omnivoice.refText, '테스트 기준 음성입니다.');
+  assert.equal(settings.omnivoice.language, 'ko');
+  assert.equal(settings.omnivoice.speaker, 'warm korean male voice');
+  assert.equal(settings.omnivoice.timeoutMs, 180000);
+  assert.equal(settings.omnivoice.useForProgress, true);
+});
+
+test('buildTtsSettings normalizes Qwen3 TTS CLI settings and aliases qwen3', () => {
+  const root = '/project';
+  const settings = buildTtsSettings({
+    TTS_BACKEND: 'qwen3',
+    QWEN3TTS_COMMAND: 'audio',
+    QWEN3TTS_MODE: 'clone',
+    QWEN3TTS_MODEL: 'base-8bit',
+    QWEN3TTS_LANGUAGE: 'korean',
+    QWEN3TTS_SPEAKER: 'sohee',
+    QWEN3TTS_INSTRUCT: 'calm conversational Korean',
+    QWEN3TTS_REF_AUDIO: './voice-samples/me.wav',
+    QWEN3TTS_REF_TEXT: '테스트 기준 음성입니다.',
+    QWEN3TTS_STREAM: '0',
+    QWEN3TTS_TIMEOUT_MS: '90000',
+    QWEN3TTS_PROGRESS: '1',
+  }, root);
+
+  assert.equal(settings.backend, 'qwen3tts');
+  assert.equal(settings.qwen3tts.command, 'audio');
+  assert.equal(settings.qwen3tts.mode, 'clone');
+  assert.equal(settings.qwen3tts.model, 'base-8bit');
+  assert.equal(settings.qwen3tts.language, 'korean');
+  assert.equal(settings.qwen3tts.speaker, 'sohee');
+  assert.equal(settings.qwen3tts.instruct, 'calm conversational Korean');
+  assert.equal(settings.qwen3tts.refAudio, path.join(root, 'voice-samples', 'me.wav'));
+  assert.equal(settings.qwen3tts.refText, '테스트 기준 음성입니다.');
+  assert.equal(settings.qwen3tts.stream, false);
+  assert.equal(settings.qwen3tts.timeoutMs, 90000);
+  assert.equal(settings.qwen3tts.useForProgress, true);
+});
+
+test('buildTtsSettings normalizes FireRedTTS-2 settings', () => {
+  const root = '/project';
+  const settings = buildTtsSettings({
+    TTS_BACKEND: 'firered',
+    FIREREDTTS2_COMMAND: './bin/fireredtts2',
+    FIREREDTTS2_PRETRAINED_DIR: './models/FireRedTTS2',
+    FIREREDTTS2_DEVICE: 'mps',
+    FIREREDTTS2_GEN_TYPE: 'monologue',
+    FIREREDTTS2_SPEAKER: 'S1',
+    FIREREDTTS2_PROMPT_AUDIO: './voice-samples/me.wav',
+    FIREREDTTS2_PROMPT_TEXT: '테스트 기준 음성입니다.',
+    FIREREDTTS2_BF16: '1',
+    FIREREDTTS2_TIMEOUT_MS: '240000',
+    FIREREDTTS2_PROGRESS: '1',
+  }, root);
+
+  assert.equal(settings.backend, 'fireredtts2');
+  assert.equal(settings.fireredtts2.command, './bin/fireredtts2');
+  assert.equal(settings.fireredtts2.pretrainedDir, path.join(root, 'models', 'FireRedTTS2'));
+  assert.equal(settings.fireredtts2.device, 'mps');
+  assert.equal(settings.fireredtts2.genType, 'monologue');
+  assert.equal(settings.fireredtts2.speaker, 'S1');
+  assert.equal(settings.fireredtts2.promptAudio, path.join(root, 'voice-samples', 'me.wav'));
+  assert.equal(settings.fireredtts2.promptText, '테스트 기준 음성입니다.');
+  assert.equal(settings.fireredtts2.useBf16, true);
+  assert.equal(settings.fireredtts2.timeoutMs, 240000);
+  assert.equal(settings.fireredtts2.useForProgress, true);
+});
+
+test('buildTtsSettings normalizes MOSS-TTS-Nano settings', () => {
+  const root = '/project';
+  const settings = buildTtsSettings({
+    TTS_BACKEND: 'moss-tts-nano',
+    MOSSTTSNANO_COMMAND: 'python3',
+    MOSSTTSNANO_SCRIPT: './vendor/MOSS-TTS-Nano/infer.py',
+    MOSSTTSNANO_CHECKPOINT: './models/MOSS-TTS-Nano',
+    MOSSTTSNANO_AUDIO_TOKENIZER: './models/MOSS-Audio-Tokenizer-Nano',
+    MOSSTTSNANO_MODE: 'voice_clone',
+    MOSSTTSNANO_LANGUAGE: 'ko',
+    MOSSTTSNANO_DEVICE: 'cpu',
+    MOSSTTSNANO_DTYPE: 'float32',
+    MOSSTTSNANO_PROMPT_AUDIO: './voice-samples/me.wav',
+    MOSSTTSNANO_PROMPT_TEXT: '테스트 기준 음성입니다.',
+    MOSSTTSNANO_MAX_NEW_FRAMES: '256',
+    MOSSTTSNANO_SEED: '7',
+    MOSSTTSNANO_TIMEOUT_MS: '90000',
+    MOSSTTSNANO_PROGRESS: '1',
+  }, root);
+
+  assert.equal(settings.backend, 'mossttsnano');
+  assert.equal(settings.mossttsnano.command, 'python3');
+  assert.equal(settings.mossttsnano.script, path.join(root, 'vendor', 'MOSS-TTS-Nano', 'infer.py'));
+  assert.equal(settings.mossttsnano.checkpoint, './models/MOSS-TTS-Nano');
+  assert.equal(settings.mossttsnano.audioTokenizer, './models/MOSS-Audio-Tokenizer-Nano');
+  assert.equal(settings.mossttsnano.mode, 'voice_clone');
+  assert.equal(settings.mossttsnano.language, 'ko');
+  assert.equal(settings.mossttsnano.device, 'cpu');
+  assert.equal(settings.mossttsnano.dtype, 'float32');
+  assert.equal(settings.mossttsnano.promptAudio, path.join(root, 'voice-samples', 'me.wav'));
+  assert.equal(settings.mossttsnano.promptText, '테스트 기준 음성입니다.');
+  assert.equal(settings.mossttsnano.maxNewFrames, 256);
+  assert.equal(settings.mossttsnano.seed, '7');
+  assert.equal(settings.mossttsnano.timeoutMs, 90000);
+  assert.equal(settings.mossttsnano.useForProgress, true);
+});
+
+test('buildTtsSettings normalizes MOSS-TTS-Nano MLX hybrid settings', () => {
+  const root = '/project';
+  const settings = buildTtsSettings({
+    TTS_BACKEND: 'moss-mlx',
+    MOSSTTSNANO_COMMAND: 'python3',
+    MOSSTTSNANO_MLX_SCRIPT: './integrations/mossttsnano_mlx/synth.py',
+    MOSSTTSNANO_MLX_WORKER_SCRIPT: './integrations/mossttsnano_mlx/worker.py',
+    MOSSTTSNANO_MLX_WORKER: '1',
+    MOSSTTSNANO_MLX_WORKER_STARTUP_TIMEOUT_MS: '240000',
+    MOSSTTSNANO_SCRIPT: './vendor/MOSS-TTS-Nano/infer.py',
+    MOSSTTSNANO_CHECKPOINT: './models/MOSS-TTS-Nano',
+    MOSSTTSNANO_AUDIO_TOKENIZER: './models/MOSS-Audio-Tokenizer-Nano',
+    MOSSTTSNANO_MODE: 'voice_clone',
+    MOSSTTSNANO_LANGUAGE: 'ko',
+    MOSSTTSNANO_TORCH_DEVICE: 'cpu',
+    MOSSTTSNANO_TORCH_DTYPE: 'float32',
+    MOSSTTSNANO_PROMPT_AUDIO: './voice-samples/me.wav',
+    MOSSTTSNANO_PROMPT_TEXT: '테스트 기준 음성입니다.',
+    MOSSTTSNANO_MAX_NEW_FRAMES: '120',
+    MOSSTTSNANO_SEED: '7',
+    MOSSTTSNANO_MLX_TIMEOUT_MS: '180000',
+    MOSSTTSNANO_MLX_PROGRESS: '0',
+  }, root);
+
+  assert.equal(settings.backend, 'mossttsnano_mlx');
+  assert.equal(settings.mossttsnano_mlx.python, 'python3');
+  assert.equal(settings.mossttsnano_mlx.script, path.join(root, 'integrations', 'mossttsnano_mlx', 'synth.py'));
+  assert.equal(settings.mossttsnano_mlx.workerScript, path.join(root, 'integrations', 'mossttsnano_mlx', 'worker.py'));
+  assert.equal(settings.mossttsnano_mlx.workerEnabled, true);
+  assert.equal(settings.mossttsnano_mlx.workerStartupTimeoutMs, 240000);
+  assert.equal(settings.mossttsnano_mlx.torchInferScript, path.join(root, 'vendor', 'MOSS-TTS-Nano', 'infer.py'));
+  assert.equal(settings.mossttsnano_mlx.checkpoint, './models/MOSS-TTS-Nano');
+  assert.equal(settings.mossttsnano_mlx.audioTokenizer, './models/MOSS-Audio-Tokenizer-Nano');
+  assert.equal(settings.mossttsnano_mlx.mode, 'voice_clone');
+  assert.equal(settings.mossttsnano_mlx.language, 'ko');
+  assert.equal(settings.mossttsnano_mlx.torchDevice, 'cpu');
+  assert.equal(settings.mossttsnano_mlx.torchDtype, 'float32');
+  assert.equal(settings.mossttsnano_mlx.promptAudio, path.join(root, 'voice-samples', 'me.wav'));
+  assert.equal(settings.mossttsnano_mlx.promptText, '테스트 기준 음성입니다.');
+  assert.equal(settings.mossttsnano_mlx.maxNewFrames, 120);
+  assert.equal(settings.mossttsnano_mlx.seed, '7');
+  assert.equal(settings.mossttsnano_mlx.timeoutMs, 180000);
+  assert.equal(settings.mossttsnano_mlx.useForProgress, false);
+});
+
+test('buildTtsSettings normalizes NeuTTS Air settings and aliases neutts air', () => {
+  const root = '/project';
+  const settings = buildTtsSettings({
+    TTS_BACKEND: 'neutts-air',
+    NEUTTSAIR_PYTHON: './.venv-neuttsair/bin/python',
+    NEUTTSAIR_SCRIPT: './integrations/neuttsair/synth.py',
+    NEUTTSAIR_BACKBONE_REPO: 'neuphonic/neutts-air-q4-gguf',
+    NEUTTSAIR_BACKBONE_DEVICE: 'mps',
+    NEUTTSAIR_CODEC_REPO: 'neuphonic/neucodec',
+    NEUTTSAIR_CODEC_DEVICE: 'mps',
+    NEUTTSAIR_REF_AUDIO: './voice-samples/me.wav',
+    NEUTTSAIR_REF_TEXT: 'Reference voice text.',
+    NEUTTSAIR_LANGUAGE: 'en',
+    NEUTTSAIR_SAMPLE_RATE: '24000',
+    NEUTTSAIR_TIMEOUT_MS: '120000',
+    NEUTTSAIR_PROGRESS: '1',
+  }, root);
+
+  assert.equal(settings.backend, 'neuttsair');
+  assert.equal(settings.neuttsair.python, path.join(root, '.venv-neuttsair', 'bin', 'python'));
+  assert.equal(settings.neuttsair.script, path.join(root, 'integrations', 'neuttsair', 'synth.py'));
+  assert.equal(settings.neuttsair.backboneRepo, 'neuphonic/neutts-air-q4-gguf');
+  assert.equal(settings.neuttsair.backboneDevice, 'mps');
+  assert.equal(settings.neuttsair.codecRepo, 'neuphonic/neucodec');
+  assert.equal(settings.neuttsair.codecDevice, 'mps');
+  assert.equal(settings.neuttsair.refAudio, path.join(root, 'voice-samples', 'me.wav'));
+  assert.equal(settings.neuttsair.refText, 'Reference voice text.');
+  assert.equal(settings.neuttsair.language, 'en');
+  assert.equal(settings.neuttsair.sampleRate, 24000);
+  assert.equal(settings.neuttsair.timeoutMs, 120000);
+  assert.equal(settings.neuttsair.useForProgress, true);
+});
+
 test('buildTtsSettings falls back to edge for unsupported backend', () => {
   const settings = buildTtsSettings({ TTS_BACKEND: 'unknown' }, '/project');
   assert.equal(settings.backend, 'edge');

package/app-node/tts_voice_config.mjs

@@ -34,6 +34,55 @@ export const DEFAULT_TTS_VOICE_CONFIG = {
         m1: { label: 'Supertonic M1', language: 'ko', voice: 'M1' },
       },
     },
+    omnivoice: {
+      currentVoiceType: 'cloned_reference',
+      voices: {
+        cloned_reference: { label: 'OmniVoice reference sample', language: 'ko', voice: 'voice-samples/user-reference.wav' },
+        designed_speaker: { label: 'OmniVoice designed speaker', language: 'ko', voice: 'warm korean male voice' },
+      },
+    },
+    qwen3tts: {
+      currentVoiceType: 'korean_preset',
+      voices: {
+        korean_preset: { label: 'Qwen3 TTS Korean preset', language: 'ko', voice: 'sohee' },
+        cloned_reference: { label: 'Qwen3 TTS reference sample', language: 'ko', voice: 'voice-samples/user-reference.wav' },
+        designed_speaker: { label: 'Qwen3 TTS designed speaker', language: 'ko', voice: 'calm conversational Korean voice' },
+      },
+    },
+    mlxaudio: {
+      currentVoiceType: 'qwen3_mlx',
+      voices: {
+        qwen3_mlx: { label: 'MLX Audio Qwen3 speaker', language: 'ko', voice: 'Chelsie' },
+      },
+    },
+    neuttsair: {
+      currentVoiceType: 'cloned_reference',
+      voices: {
+        cloned_reference: { label: 'NeuTTS Air reference sample', language: 'en', voice: 'voice-samples/user-reference.wav' },
+        default_sample: { label: 'NeuTTS Air bundled sample', language: 'en', voice: 'vendor/neutts-air/samples/jo.wav' },
+      },
+    },
+    fireredtts2: {
+      currentVoiceType: 'prompt_reference',
+      voices: {
+        prompt_reference: { label: 'FireRedTTS-2 prompt reference', language: 'ko', voice: 'voice-samples/user-reference.wav' },
+        random_speaker: { label: 'FireRedTTS-2 random speaker', language: 'ko', voice: '' },
+      },
+    },
+    mossttsnano: {
+      currentVoiceType: 'prompt_reference',
+      voices: {
+        prompt_reference: { label: 'MOSS-TTS-Nano prompt reference', language: 'ko', voice: 'voice-samples/user-reference.wav' },
+        continuation: { label: 'MOSS-TTS-Nano continuation/default', language: 'ko', voice: '' },
+      },
+    },
+    mossttsnano_mlx: {
+      currentVoiceType: 'prompt_reference',
+      voices: {
+        prompt_reference: { label: 'MOSS-TTS-Nano MLX hybrid prompt reference', language: 'ko', voice: 'voice-samples/user-reference.wav' },
+        continuation: { label: 'MOSS-TTS-Nano MLX hybrid continuation/default', language: 'ko', voice: '' },
+      },
+    },
   },
 };
 
@@ -47,7 +96,37 @@ export function defaultTtsVoiceConfig() {
 
 function normalizeBackend(value, config) {
   const key = String(value || '').trim().toLowerCase();
-  return config.backends?.[key] ? key : 'edge';
+  const aliases = new Map([
+    ['q3', 'qwen3tts'],
+    ['qwen3', 'qwen3tts'],
+    ['qwen3-tts', 'qwen3tts'],
+    ['qtts', 'qwen3tts'],
+    ['qwen3-mlx', 'mlxaudio'],
+    ['mlx', 'mlxaudio'],
+    ['mlx-audio', 'mlxaudio'],
+    ['neutts', 'neuttsair'],
+    ['neutts-air', 'neuttsair'],
+    ['neutts air', 'neuttsair'],
+    ['neuttsair', 'neuttsair'],
+    ['neu-tts-air', 'neuttsair'],
+    ['neu tts air', 'neuttsair'],
+    ['firered', 'fireredtts2'],
+    ['fireredtts', 'fireredtts2'],
+    ['firered-tts-2', 'fireredtts2'],
+    ['fireredtts-2', 'fireredtts2'],
+    ['moss', 'mossttsnano'],
+    ['moss-tts', 'mossttsnano'],
+    ['mossnano', 'mossttsnano'],
+    ['moss-tts-nano', 'mossttsnano'],
+    ['openmoss', 'mossttsnano'],
+    ['moss-mlx', 'mossttsnano_mlx'],
+    ['moss mlx', 'mossttsnano_mlx'],
+    ['mossttsnano-mlx', 'mossttsnano_mlx'],
+    ['mossttsnano_mlx', 'mossttsnano_mlx'],
+    ['openmoss-mlx', 'mossttsnano_mlx'],
+  ]);
+  const normalized = aliases.get(key) || key;
+  return config.backends?.[normalized] ? normalized : 'edge';
 }
 
 function normalizeVoiceType(backendConfig, requested) {
@@ -75,7 +154,10 @@ export function effectiveTtsVoiceSelection(config, env = {}) {
   const merged = config || defaultTtsVoiceConfig();
   const backend = normalizeBackend(env.TTS_BACKEND || merged.currentBackend, merged);
   const backendConfig = merged.backends[backend] || merged.backends.edge;
-  const voiceType = normalizeVoiceType(backendConfig, env.TTS_VOICE_TYPE || merged.currentVoiceType || backendConfig.currentVoiceType);
+  const requestedVoiceType = env.TTS_VOICE_TYPE || merged.currentVoiceType;
+  const voiceType = backendConfig.voices?.[requestedVoiceType]
+    ? requestedVoiceType
+    : normalizeVoiceType(backendConfig, backendConfig.currentVoiceType);
   const voice = backendConfig.voices[voiceType];
   return { backend, voiceType, voice, backendConfig };
 }
@@ -83,6 +165,41 @@ export function effectiveTtsVoiceSelection(config, env = {}) {
 export function applyTtsVoiceSelectionToEnv(env = {}, selection) {
   const next = { ...env, TTS_BACKEND: selection.backend, TTS_VOICE_TYPE: selection.voiceType };
   if (selection.backend === 'edge') next.TTS_VOICE = selection.voice.voice;
+  if (selection.backend === 'qwen3tts') {
+    if (selection.voiceType === 'cloned_reference') {
+      next.QWEN3TTS_MODE = 'clone';
+      next.QWEN3TTS_REF_AUDIO = selection.voice.voice;
+    } else if (selection.voiceType === 'designed_speaker') {
+      next.QWEN3TTS_MODE = 'design';
+      next.QWEN3TTS_INSTRUCT = selection.voice.voice;
+    } else {
+      next.QWEN3TTS_MODE = 'custom';
+      next.QWEN3TTS_SPEAKER = selection.voice.voice;
+    }
+  }
+  if (selection.backend === 'mlxaudio') {
+    if (selection.voice?.voice) next.MLXAUDIO_VOICE = selection.voice.voice;
+  }
+  if (selection.backend === 'neuttsair') {
+    if (selection.voice?.voice) next.NEUTTSAIR_REF_AUDIO = selection.voice.voice;
+  }
+  if (selection.backend === 'fireredtts2') {
+    if (selection.voice?.voice) next.FIREREDTTS2_PROMPT_AUDIO = selection.voice.voice;
+  }
+  if (selection.backend === 'mossttsnano') {
+    if (selection.voiceType === 'continuation') next.MOSSTTSNANO_MODE = 'continuation';
+    else {
+      next.MOSSTTSNANO_MODE = 'voice_clone';
+      if (selection.voice?.voice) next.MOSSTTSNANO_PROMPT_AUDIO = selection.voice.voice;
+    }
+  }
+  if (selection.backend === 'mossttsnano_mlx') {
+    if (selection.voiceType === 'continuation') next.MOSSTTSNANO_MODE = 'continuation';
+    else {
+      next.MOSSTTSNANO_MODE = 'voice_clone';
+      if (selection.voice?.voice) next.MOSSTTSNANO_PROMPT_AUDIO = selection.voice.voice;
+    }
+  }
   if (selection.voice?.language) next.VOICE_LANGUAGE = selection.voice.language;
   return next;
 }
@@ -113,6 +230,23 @@ export function voiceCommandFromTranscript(text) {
   const raw = String(text || '').trim();
   if (!raw) return null;
   const compact = raw.toLowerCase().replace(/\s+/g, '');
+  const looksLikeBackend = /\b(tts|voice|speech|audio)\b.*\bbackend\b|\bbackend\b.*\b(tts|voice|speech|audio)\b/i.test(raw)
+    || /(tts|음성|목소리).*(백엔드|백앤드|backend).*(바꿔|변경|설정|해줘|로)/iu.test(raw)
+    || /(백엔드|백앤드|backend).*(옴니보이스|오픈보이스|엣지|수퍼토닉|슈퍼토닉|스피치스위프트|큐원|큐웬|qwen|q3|qtts|firered|moss|openmoss|neutts|neu\s*tts|뉴티티에스|뉴티TS|speechswift|omnivoice|openvoice|edge|supertonic)/iu.test(raw)
+    || /tts를.*(옴니보이스|오픈보이스|엣지|수퍼토닉|슈퍼토닉|스피치스위프트|큐원|큐웬|qwen|q3|qtts|firered|moss|openmoss|neutts|neu\s*tts|뉴티티에스|뉴티TS|omnivoice|openvoice|edge|supertonic|speechswift).*바꿔/iu.test(raw);
+  if (looksLikeBackend) {
+    if (/(neutts\s*-?\s*air|neu\s*tts\s*-?\s*air|neuttsair|뉴\s*티\s*티\s*에스\s*에어|뉴티티에스\s*에어|뉴티TS\s*에어)/iu.test(raw)) return { backend: 'neuttsair' };
+    if (/(omnivoice|omni voice|옴니보이스|업니보이스|옴니|업니)/iu.test(raw)) return { backend: 'omnivoice' };
+    if (/(openvoice|open voice|오픈보이스|오픈 보이스)/iu.test(raw)) return { backend: 'openvoice' };
+    if (/(speechswift|speech swift|스피치스위프트|스피치 스위프트|cosyvoice|코지보이스)/iu.test(raw)) return { backend: 'speechswift' };
+    if (/(qwen3mlx|qwen mlx|qwen3 mlx|mlx-audio|mlx audio|엠엘엑스|mlx)/iu.test(raw)) return { backend: 'mlxaudio' };
+    if (/(neutts-air|neuttsair|neutts|neu tts air|뉴티티에스|뉴티츠|뉴티에스)/iu.test(raw)) return { backend: 'neuttsair' };
+    if (/(qwen3|qwen|q3|qtts|큐원|큐웬|큐엔|큐3|큐삼)/iu.test(raw)) return { backend: 'qwen3tts' };
+    if (/(fireredtts2|fireredtts|firered|fire red|파이어레드)/iu.test(raw)) return { backend: 'fireredtts2' };
+    if (/(moss-tts-nano|moss tts nano|mossnano|moss|openmoss|모스|오픈모스)/iu.test(raw)) return { backend: 'mossttsnano' };
+    if (/(supertonic|수퍼토닉|슈퍼토닉)/iu.test(raw)) return { backend: 'supertonic' };
+    if (/(edge|엣지)/iu.test(raw)) return { backend: 'edge' };
+  }
   const looksLikeVoice = /\b(change|switch|set)\b.*\b(voice|speaker)\b/i.test(raw)
     || /\b(voice|speaker)\b.*\b(to|as)\b/i.test(raw)
     || /(목소리|음성).*(바꿔|변경|설정|해줘)|목소리.*로|음성.*로/u.test(compact);

package/app-node/tts_voice_config.test.mjs

@@ -15,6 +15,11 @@ import {
   writeTtsVoiceConfig,
 } from './tts_voice_config.mjs';
 
+const __tempRoots = [];
+test.after(() => {
+  for (const root of __tempRoots) try { fs.rmSync(root, { recursive: true, force: true }); } catch {}
+});
+
 test('effectiveTtsVoiceSelection reads backend and voice type from config', () => {
   const config = defaultTtsVoiceConfig();
   config.currentBackend = 'edge';
@@ -27,6 +32,26 @@ test('effectiveTtsVoiceSelection reads backend and voice type from config', () =
   assert.equal(selected.voice.voice, 'ko-KR-SunHiNeural');
 });
 
+test('effectiveTtsVoiceSelection falls back to backend voice when env voice type belongs to another backend', () => {
+  const config = defaultTtsVoiceConfig();
+  config.currentBackend = 'edge';
+  config.currentVoiceType = 'korean_male';
+  config.backends.edge.currentVoiceType = 'korean_male';
+
+  const selected = effectiveTtsVoiceSelection(config, { TTS_BACKEND: 'edge', TTS_VOICE_TYPE: 'cloned_reference' });
+
+  assert.equal(selected.backend, 'edge');
+  assert.equal(selected.voiceType, 'korean_male');
+  assert.equal(selected.voice.voice, 'ko-KR-InJoonNeural');
+});
+
+test('effectiveTtsVoiceSelection accepts Qwen3 backend aliases from env', () => {
+  const selected = effectiveTtsVoiceSelection(defaultTtsVoiceConfig(), { TTS_BACKEND: 'qwen3' });
+
+  assert.equal(selected.backend, 'qwen3tts');
+  assert.equal(selected.voiceType, 'korean_preset');
+});
+
 test('applyTtsVoiceSelectionToEnv updates backend voice and voice language together', () => {
   const selected = effectiveTtsVoiceSelection(updateTtsVoiceConfig(defaultTtsVoiceConfig(), { voiceType: 'korean_male' }), {});
 
@@ -52,8 +77,77 @@ test('voiceCommandFromTranscript detects voice type changes', () => {
   assert.equal(voiceCommandFromTranscript('change language to Korean'), null);
 });
 
+test('voiceCommandFromTranscript detects TTS backend changes', () => {
+  assert.deepEqual(voiceCommandFromTranscript('change TTS backend to OmniVoice'), { backend: 'omnivoice' });
+  assert.deepEqual(voiceCommandFromTranscript('음성 백엔드 옴니보이스로 바꿔'), { backend: 'omnivoice' });
+  assert.deepEqual(voiceCommandFromTranscript('TTS를 Edge로 바꿔'), { backend: 'edge' });
+  assert.deepEqual(voiceCommandFromTranscript('TTS를 qwen3로 바꿔'), { backend: 'qwen3tts' });
+  assert.deepEqual(voiceCommandFromTranscript('음성 백엔드 큐웬으로 바꿔'), { backend: 'qwen3tts' });
+  assert.deepEqual(voiceCommandFromTranscript('TTS backend to FireRedTTS-2'), { backend: 'fireredtts2' });
+  assert.deepEqual(voiceCommandFromTranscript('음성 백엔드 모스로 바꿔'), { backend: 'mossttsnano' });
+  assert.deepEqual(voiceCommandFromTranscript('TTS backend to NeuTTS Air'), { backend: 'neuttsair' });
+  assert.deepEqual(voiceCommandFromTranscript('음성 백엔드 뉴티티에스 에어로 바꿔'), { backend: 'neuttsair' });
+});
+
+test('applyTtsVoiceSelectionToEnv maps Qwen3 voice types to CLI mode env', () => {
+  const preset = effectiveTtsVoiceSelection(updateTtsVoiceConfig(defaultTtsVoiceConfig(), { backend: 'qwen3tts', voiceType: 'korean_preset' }), {});
+  assert.deepEqual(applyTtsVoiceSelectionToEnv({}, preset), {
+    TTS_BACKEND: 'qwen3tts',
+    TTS_VOICE_TYPE: 'korean_preset',
+    QWEN3TTS_MODE: 'custom',
+    QWEN3TTS_SPEAKER: 'sohee',
+    VOICE_LANGUAGE: 'ko',
+  });
+
+  const clone = effectiveTtsVoiceSelection(updateTtsVoiceConfig(defaultTtsVoiceConfig(), { backend: 'qwen3tts', voiceType: 'cloned_reference' }), {});
+  assert.deepEqual(applyTtsVoiceSelectionToEnv({}, clone), {
+    TTS_BACKEND: 'qwen3tts',
+    TTS_VOICE_TYPE: 'cloned_reference',
+    QWEN3TTS_MODE: 'clone',
+    QWEN3TTS_REF_AUDIO: 'voice-samples/user-reference.wav',
+    VOICE_LANGUAGE: 'ko',
+  });
+});
+
+test('applyTtsVoiceSelectionToEnv maps FireRedTTS-2 and MOSS prompt references', () => {
+  const fire = effectiveTtsVoiceSelection(updateTtsVoiceConfig(defaultTtsVoiceConfig(), { backend: 'fireredtts2', voiceType: 'prompt_reference' }), {});
+  assert.deepEqual(applyTtsVoiceSelectionToEnv({}, fire), {
+    TTS_BACKEND: 'fireredtts2',
+    TTS_VOICE_TYPE: 'prompt_reference',
+    FIREREDTTS2_PROMPT_AUDIO: 'voice-samples/user-reference.wav',
+    VOICE_LANGUAGE: 'ko',
+  });
+
+  const moss = effectiveTtsVoiceSelection(updateTtsVoiceConfig(defaultTtsVoiceConfig(), { backend: 'mossttsnano', voiceType: 'prompt_reference' }), {});
+  assert.deepEqual(applyTtsVoiceSelectionToEnv({}, moss), {
+    TTS_BACKEND: 'mossttsnano',
+    TTS_VOICE_TYPE: 'prompt_reference',
+    MOSSTTSNANO_MODE: 'voice_clone',
+    MOSSTTSNANO_PROMPT_AUDIO: 'voice-samples/user-reference.wav',
+    VOICE_LANGUAGE: 'ko',
+  });
+
+  const neutts = effectiveTtsVoiceSelection(updateTtsVoiceConfig(defaultTtsVoiceConfig(), { backend: 'neuttsair', voiceType: 'cloned_reference' }), {});
+  assert.deepEqual(applyTtsVoiceSelectionToEnv({}, neutts), {
+    TTS_BACKEND: 'neuttsair',
+    TTS_VOICE_TYPE: 'cloned_reference',
+    NEUTTSAIR_REF_AUDIO: 'voice-samples/user-reference.wav',
+    VOICE_LANGUAGE: 'en',
+  });
+});
+
+test('updateTtsVoiceConfig can switch to OmniVoice backend default voice', () => {
+  const config = updateTtsVoiceConfig(defaultTtsVoiceConfig(), { backend: 'omnivoice' });
+  const selected = effectiveTtsVoiceSelection(config, {});
+
+  assert.equal(selected.backend, 'omnivoice');
+  assert.equal(selected.voiceType, 'cloned_reference');
+  assert.equal(selected.voice.voice, 'voice-samples/user-reference.wav');
+});
+
 test('read and write voice config round trips current selection', () => {
   const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'voice-config-test-'));
+  __tempRoots.push(dir);
   const file = path.join(dir, 'tts-voices.json');
   const config = updateTtsVoiceConfig(defaultTtsVoiceConfig(), { voiceType: 'korean_female' });