verbalcoding 0.2.12 → 0.2.13

package/app-node/tts_backends.mjs

@@ -1,3 +1,4 @@
+import { spawn as spawnProcess } from 'node:child_process';
 import fs from 'node:fs';
 import os from 'node:os';
 import path from 'node:path';
@@ -17,6 +18,11 @@ function execOptions(base, signal) {
   return signal ? { ...base, signal } : base;
 }
 
+export function notifyTtsFallback(deps, backend, error, kind) {
+  (deps.warn || (() => {}))(`${backend} failed; falling back to edge`, error?.message || error);
+  try { deps.onFallback?.({ backend, error, kind }); } catch {}
+}
+
 function openVoicePython(openvoice, existsSync = fs.existsSync) {
   const venvPython = path.join(openvoice.venv, 'bin', 'python');
   if (existsSync(venvPython)) return venvPython;
@@ -56,6 +62,119 @@ function supertonicEnv(baseEnv, supertonic) {
   return env;
 }
 
+function omniVoiceArgs(text, out, omnivoice) {
+  const args = [
+    path.resolve(path.dirname(new URL(import.meta.url).pathname), '..', 'integrations', 'omnivoice', 'synth.py'),
+    '--text', text,
+    '--output', out,
+    '--model', omnivoice.model,
+    '--device', omnivoice.device,
+    '--dtype', omnivoice.dtype,
+  ];
+  if (omnivoice.refAudio) args.push('--ref-audio', omnivoice.refAudio);
+  if (omnivoice.refText) args.push('--ref-text', omnivoice.refText);
+  if (omnivoice.language) args.push('--language', omnivoice.language);
+  if (omnivoice.speaker) args.push('--speaker', omnivoice.speaker);
+  return args;
+}
+
+function qwen3TtsArgs(text, out, qwen3tts) {
+  const args = ['speak', text, '--engine', 'qwen3', '--output', out];
+  if (qwen3tts.language) args.push('--language', qwen3tts.language);
+  if (qwen3tts.stream) args.push('--stream');
+  if (qwen3tts.mode === 'clone') {
+    args.push('--model', qwen3tts.model || 'base');
+    if (qwen3tts.refAudio) args.push('--voice-sample', qwen3tts.refAudio);
+  } else if (qwen3tts.mode === 'design') {
+    args.push('--model', qwen3tts.model || 'customVoice');
+    if (qwen3tts.instruct) args.push('--instruct', qwen3tts.instruct);
+  } else {
+    args.push('--model', qwen3tts.model || 'customVoice');
+    if (qwen3tts.speaker) args.push('--speaker', qwen3tts.speaker);
+    if (qwen3tts.instruct) args.push('--instruct', qwen3tts.instruct);
+  }
+  return args;
+}
+
+function mlxAudioArgs(text, out, mlxaudio) {
+  const args = [
+    path.resolve(path.dirname(new URL(import.meta.url).pathname), '..', 'integrations', 'mlxaudio', 'synth.py'),
+    '--text', text,
+    '--output', out,
+    '--model', mlxaudio.model,
+    '--voice', mlxaudio.voice,
+  ];
+  if (mlxaudio.langCode) args.push('--lang-code', mlxaudio.langCode);
+  if (mlxaudio.stream) args.push('--stream');
+  return args;
+}
+
+function neuTtsAirArgs(text, out, neuttsair) {
+  const args = [
+    neuttsair.script,
+    '--text', text,
+    '--output', out,
+    '--backbone-repo', neuttsair.backboneRepo,
+    '--codec-repo', neuttsair.codecRepo,
+    '--backbone-device', neuttsair.backboneDevice,
+    '--codec-device', neuttsair.codecDevice,
+    '--ref-audio', neuttsair.refAudio,
+    '--language', neuttsair.language,
+    '--sample-rate', String(neuttsair.sampleRate),
+  ];
+  if (neuttsair.refText) args.push('--ref-text', neuttsair.refText);
+  if (neuttsair.refTextFile) args.push('--ref-text-file', neuttsair.refTextFile);
+  if (neuttsair.cacheRef) args.push('--cache-ref');
+  return args;
+}
+
+function fireRedTts2Args(text, out, fireredtts2) {
+  const args = ['--text', text, '--output', out];
+  if (fireredtts2.pretrainedDir) args.push('--pretrained-dir', fireredtts2.pretrainedDir);
+  if (fireredtts2.device) args.push('--device', fireredtts2.device);
+  if (fireredtts2.genType) args.push('--gen-type', fireredtts2.genType);
+  if (fireredtts2.speaker) args.push('--speaker', fireredtts2.speaker);
+  if (fireredtts2.promptAudio) args.push('--prompt-audio', fireredtts2.promptAudio);
+  if (fireredtts2.promptText) args.push('--prompt-text', fireredtts2.promptText);
+  if (fireredtts2.useBf16) args.push('--bf16');
+  return args;
+}
+
+function mossTtsNanoArgs(text, out, mossttsnano) {
+  const args = [mossttsnano.script || 'infer.py', '--text', text, '--output-audio-path', out];
+  if (mossttsnano.checkpoint) args.push('--checkpoint', mossttsnano.checkpoint);
+  if (mossttsnano.audioTokenizer) args.push('--audio-tokenizer-pretrained-name-or-path', mossttsnano.audioTokenizer);
+  if (mossttsnano.mode) args.push('--mode', mossttsnano.mode);
+  if (mossttsnano.device) args.push('--device', mossttsnano.device);
+  if (mossttsnano.dtype) args.push('--dtype', mossttsnano.dtype);
+  if (mossttsnano.promptAudio) args.push('--prompt-audio-path', mossttsnano.promptAudio);
+  if (mossttsnano.promptText) args.push('--prompt-text', mossttsnano.promptText);
+  if (mossttsnano.maxNewFrames) args.push('--max-new-frames', String(mossttsnano.maxNewFrames));
+  if (mossttsnano.seed) args.push('--seed', String(mossttsnano.seed));
+  if (mossttsnano.disableWetext !== false) args.push('--disable-wetext-processing');
+  return args;
+}
+
+function mossTtsNanoMlxArgs(text, out, mossttsnanoMlx) {
+  const args = [
+    mossttsnanoMlx.script,
+    '--text', text,
+    '--output-audio-path', out,
+    '--checkpoint', mossttsnanoMlx.checkpoint,
+    '--audio-tokenizer-pretrained-name-or-path', mossttsnanoMlx.audioTokenizer,
+    '--mode', mossttsnanoMlx.mode,
+    '--torch-infer-script', mossttsnanoMlx.torchInferScript,
+    '--torch-device', mossttsnanoMlx.torchDevice,
+    '--torch-dtype', mossttsnanoMlx.torchDtype,
+    '--max-new-frames', String(mossttsnanoMlx.maxNewFrames),
+    '--disable-wetext-processing',
+  ];
+  if (mossttsnanoMlx.promptAudio) args.push('--prompt-audio-path', mossttsnanoMlx.promptAudio);
+  if (mossttsnanoMlx.promptText) args.push('--prompt-text', mossttsnanoMlx.promptText);
+  if (mossttsnanoMlx.seed) args.push('--seed', String(mossttsnanoMlx.seed));
+  return args;
+}
+
 async function speechSwiftServerRequest({ fetchImpl, speechswift, text, signal }) {
   const controller = new AbortController();
   const timeout = setTimeout(() => controller.abort(), speechswift.timeoutMs);
@@ -153,7 +272,7 @@ export function createOpenVoiceBackend(settings, deps = {}) {
         return validateOutput(out, fsApi);
       } catch (error) {
         fs.rm(out, { force: true }, () => {});
-        warn('openvoice failed; falling back to edge', error?.message || error);
+        notifyTtsFallback(deps, 'openvoice', error, kind);
         return edge.synthesize(text, { signal, kind });
       }
     },
@@ -198,7 +317,7 @@ export function createSpeechSwiftBackend(settings, deps = {}) {
         return validateOutput(out, fsApi);
       } catch (error) {
         fs.rm(out, { force: true }, () => {});
-        warn('speech-swift failed; falling back to edge', error?.message || error);
+        notifyTtsFallback(deps, 'speech-swift', error, kind);
         return edge.synthesize(text, { signal, kind });
       }
     },
@@ -236,16 +355,431 @@ export function createSupertonicBackend(settings, deps = {}) {
         return validateOutput(out, fsApi);
       } catch (error) {
         fs.rm(out, { force: true }, () => {});
-        warn('supertonic failed; falling back to edge', error?.message || error);
+        notifyTtsFallback(deps, 'supertonic', error, kind);
+        return edge.synthesize(text, { signal, kind });
+      }
+    },
+  };
+}
+
+export function createOmniVoiceBackend(settings, deps = {}) {
+  const execFileAsync = deps.execFileAsync;
+  if (!execFileAsync) throw new Error('execFileAsync dependency is required');
+  const tmpdir = deps.tmpdir || os.tmpdir();
+  const warn = deps.warn || (() => {});
+  const fsApi = {
+    existsSync: deps.existsSync || fs.existsSync,
+    statSync: deps.statSync || fs.statSync,
+  };
+  const edge = createEdgeTtsBackend(settings, deps);
+  const omnivoice = settings.omnivoice;
+  return {
+    name: 'omnivoice',
+    outputExtension: omnivoice.useForProgress ? 'wav' : 'mp3',
+    cacheKeyParts() {
+      return ['omnivoice', omnivoice.model, omnivoice.device, omnivoice.dtype, omnivoice.refAudio, omnivoice.refText, omnivoice.language, omnivoice.speaker];
+    },
+    async synthesize(text, { signal, kind = 'final' } = {}) {
+      if (kind === 'progress' && !omnivoice.useForProgress) {
+        return edge.synthesize(text, { signal, kind });
+      }
+      const out = uniquePath(tmpdir, 'verbalcoding-omnivoice', 'wav');
+      try {
+        await execFileAsync(omnivoice.python || 'python3', omniVoiceArgs(text, out, omnivoice), execOptions({
+          timeout: omnivoice.timeoutMs,
+          maxBuffer: 4 * 1024 * 1024,
+        }, signal));
+        return validateOutput(out, fsApi);
+      } catch (error) {
+        fs.rm(out, { force: true }, () => {});
+        notifyTtsFallback(deps, 'omnivoice', error, kind);
+        return edge.synthesize(text, { signal, kind });
+      }
+    },
+  };
+}
+
+export function createQwen3TtsBackend(settings, deps = {}) {
+  const execFileAsync = deps.execFileAsync;
+  if (!execFileAsync) throw new Error('execFileAsync dependency is required');
+  const tmpdir = deps.tmpdir || os.tmpdir();
+  const warn = deps.warn || (() => {});
+  const fsApi = {
+    existsSync: deps.existsSync || fs.existsSync,
+    statSync: deps.statSync || fs.statSync,
+  };
+  const edge = createEdgeTtsBackend(settings, deps);
+  const qwen3tts = settings.qwen3tts;
+  return {
+    name: 'qwen3tts',
+    outputExtension: qwen3tts.useForProgress ? 'mp3' : 'mp3',
+    cacheKeyParts() {
+      return ['qwen3tts', qwen3tts.command, qwen3tts.mode, qwen3tts.language, qwen3tts.speaker, qwen3tts.instruct, qwen3tts.refAudio, qwen3tts.refText];
+    },
+    async synthesize(text, { signal, kind = 'final' } = {}) {
+      if (kind === 'progress' && !qwen3tts.useForProgress) {
+        return edge.synthesize(text, { signal, kind });
+      }
+      const out = uniquePath(tmpdir, 'verbalcoding-qwen3tts', 'mp3');
+      try {
+        await execFileAsync(qwen3tts.command, qwen3TtsArgs(text, out, qwen3tts), execOptions({
+          timeout: qwen3tts.timeoutMs,
+          maxBuffer: 4 * 1024 * 1024,
+        }, signal));
+        return validateOutput(out, fsApi);
+      } catch (error) {
+        fs.rm(out, { force: true }, () => {});
+        notifyTtsFallback(deps, 'qwen3tts', error, kind);
+        return edge.synthesize(text, { signal, kind });
+      }
+    },
+  };
+}
+
+export function createMlxAudioBackend(settings, deps = {}) {
+  const execFileAsync = deps.execFileAsync;
+  if (!execFileAsync) throw new Error('execFileAsync dependency is required');
+  const tmpdir = deps.tmpdir || os.tmpdir();
+  const warn = deps.warn || (() => {});
+  const fsApi = {
+    existsSync: deps.existsSync || fs.existsSync,
+    statSync: deps.statSync || fs.statSync,
+  };
+  const edge = createEdgeTtsBackend(settings, deps);
+  const mlxaudio = settings.mlxaudio;
+  return {
+    name: 'mlxaudio',
+    outputExtension: mlxaudio.useForProgress ? 'wav' : 'wav',
+    cacheKeyParts() {
+      return ['mlxaudio', mlxaudio.python, mlxaudio.model, mlxaudio.voice, mlxaudio.langCode, mlxaudio.stream];
+    },
+    async synthesize(text, { signal, kind = 'final' } = {}) {
+      if (kind === 'progress' && !mlxaudio.useForProgress) {
+        return edge.synthesize(text, { signal, kind });
+      }
+      const out = uniquePath(tmpdir, 'verbalcoding-mlxaudio', 'wav');
+      try {
+        await execFileAsync(mlxaudio.python || 'python3', mlxAudioArgs(text, out, mlxaudio), execOptions({
+          timeout: mlxaudio.timeoutMs,
+          maxBuffer: 4 * 1024 * 1024,
+        }, signal));
+        return validateOutput(out, fsApi);
+      } catch (error) {
+        fs.rm(out, { force: true }, () => {});
+        notifyTtsFallback(deps, 'mlxaudio', error, kind);
         return edge.synthesize(text, { signal, kind });
       }
     },
   };
 }
 
+export function createNeuTtsAirBackend(settings, deps = {}) {
+  const execFileAsync = deps.execFileAsync;
+  if (!execFileAsync) throw new Error('execFileAsync dependency is required');
+  const tmpdir = deps.tmpdir || os.tmpdir();
+  const warn = deps.warn || (() => {});
+  const fsApi = {
+    existsSync: deps.existsSync || fs.existsSync,
+    statSync: deps.statSync || fs.statSync,
+  };
+  const edge = createEdgeTtsBackend(settings, deps);
+  const neuttsair = settings.neuttsair;
+  return {
+    name: 'neuttsair',
+    outputExtension: neuttsair.useForProgress ? 'wav' : 'wav',
+    cacheKeyParts() {
+      return ['neuttsair', neuttsair.python, neuttsair.script, neuttsair.backboneRepo, neuttsair.backboneDevice, neuttsair.codecRepo, neuttsair.codecDevice, neuttsair.refAudio, neuttsair.refText, neuttsair.language, neuttsair.sampleRate];
+    },
+    async synthesize(text, { signal, kind = 'final' } = {}) {
+      if (kind === 'progress' && !neuttsair.useForProgress) {
+        return edge.synthesize(text, { signal, kind });
+      }
+      const out = uniquePath(tmpdir, 'verbalcoding-neuttsair', 'wav');
+      try {
+        await execFileAsync(neuttsair.python || 'python3', neuTtsAirArgs(text, out, neuttsair), execOptions({
+          timeout: neuttsair.timeoutMs,
+          maxBuffer: 4 * 1024 * 1024,
+        }, signal));
+        return validateOutput(out, fsApi);
+      } catch (error) {
+        fs.rm(out, { force: true }, () => {});
+        notifyTtsFallback(deps, 'neuttsair', error, kind);
+        return edge.synthesize(text, { signal, kind });
+      }
+    },
+  };
+}
+
+export function createFireRedTts2Backend(settings, deps = {}) {
+  const execFileAsync = deps.execFileAsync;
+  if (!execFileAsync) throw new Error('execFileAsync dependency is required');
+  const tmpdir = deps.tmpdir || os.tmpdir();
+  const warn = deps.warn || (() => {});
+  const fsApi = {
+    existsSync: deps.existsSync || fs.existsSync,
+    statSync: deps.statSync || fs.statSync,
+  };
+  const edge = createEdgeTtsBackend(settings, deps);
+  const fireredtts2 = settings.fireredtts2;
+  return {
+    name: 'fireredtts2',
+    outputExtension: fireredtts2.useForProgress ? 'wav' : 'wav',
+    cacheKeyParts() {
+      return ['fireredtts2', fireredtts2.command, fireredtts2.pretrainedDir, fireredtts2.device, fireredtts2.genType, fireredtts2.speaker, fireredtts2.promptAudio, fireredtts2.promptText, fireredtts2.useBf16];
+    },
+    async synthesize(text, { signal, kind = 'final' } = {}) {
+      if (kind === 'progress' && !fireredtts2.useForProgress) {
+        return edge.synthesize(text, { signal, kind });
+      }
+      const out = uniquePath(tmpdir, 'verbalcoding-fireredtts2', 'wav');
+      try {
+        await execFileAsync(fireredtts2.command, fireRedTts2Args(text, out, fireredtts2), execOptions({
+          timeout: fireredtts2.timeoutMs,
+          maxBuffer: 4 * 1024 * 1024,
+        }, signal));
+        return validateOutput(out, fsApi);
+      } catch (error) {
+        fs.rm(out, { force: true }, () => {});
+        notifyTtsFallback(deps, 'fireredtts2', error, kind);
+        return edge.synthesize(text, { signal, kind });
+      }
+    },
+  };
+}
+
+export function createMossTtsNanoBackend(settings, deps = {}) {
+  const execFileAsync = deps.execFileAsync;
+  if (!execFileAsync) throw new Error('execFileAsync dependency is required');
+  const tmpdir = deps.tmpdir || os.tmpdir();
+  const warn = deps.warn || (() => {});
+  const fsApi = {
+    existsSync: deps.existsSync || fs.existsSync,
+    statSync: deps.statSync || fs.statSync,
+  };
+  const edge = createEdgeTtsBackend(settings, deps);
+  const mossttsnano = settings.mossttsnano;
+  return {
+    name: 'mossttsnano',
+    outputExtension: mossttsnano.useForProgress ? 'wav' : 'wav',
+    cacheKeyParts() {
+      return ['mossttsnano', mossttsnano.command, mossttsnano.script, mossttsnano.checkpoint, mossttsnano.audioTokenizer, mossttsnano.mode, mossttsnano.language, mossttsnano.device, mossttsnano.dtype, mossttsnano.promptAudio, mossttsnano.promptText, mossttsnano.maxNewFrames, mossttsnano.seed];
+    },
+    async synthesize(text, { signal, kind = 'final' } = {}) {
+      if (kind === 'progress' && !mossttsnano.useForProgress) {
+        return edge.synthesize(text, { signal, kind });
+      }
+      const out = uniquePath(tmpdir, 'verbalcoding-mossttsnano', 'wav');
+      try {
+        await execFileAsync(mossttsnano.command, mossTtsNanoArgs(text, out, mossttsnano), execOptions({
+          timeout: mossttsnano.timeoutMs,
+          maxBuffer: 4 * 1024 * 1024,
+        }, signal));
+        return validateOutput(out, fsApi);
+      } catch (error) {
+        fs.rm(out, { force: true }, () => {});
+        notifyTtsFallback(deps, 'mossttsnano', error, kind);
+        return edge.synthesize(text, { signal, kind });
+      }
+    },
+  };
+}
+
+function mossTtsNanoMlxWorkerArgs(mossttsnanoMlx) {
+  const args = [
+    mossttsnanoMlx.workerScript,
+    '--checkpoint', mossttsnanoMlx.checkpoint,
+    '--audio-tokenizer-pretrained-name-or-path', mossttsnanoMlx.audioTokenizer,
+    '--mode', mossttsnanoMlx.mode,
+    '--torch-device', mossttsnanoMlx.torchDevice,
+    '--torch-dtype', mossttsnanoMlx.torchDtype,
+    '--max-new-frames', String(mossttsnanoMlx.maxNewFrames),
+  ];
+  if (mossttsnanoMlx.promptAudio) args.push('--prompt-audio-path', mossttsnanoMlx.promptAudio);
+  if (mossttsnanoMlx.promptText) args.push('--prompt-text', mossttsnanoMlx.promptText);
+  if (mossttsnanoMlx.seed) args.push('--seed', String(mossttsnanoMlx.seed));
+  return args;
+}
+
+function createJsonLineWorker({ command, args, spawn = spawnProcess, startupTimeoutMs = 120000, warn = () => {} }) {
+  let child = null;
+  let readyPromise = null;
+  let nextId = 1;
+  let stdoutBuffer = '';
+  const pending = new Map();
+
+  function rejectPending(error) {
+    for (const { reject, timer } of pending.values()) {
+      clearTimeout(timer);
+      reject(error);
+    }
+    pending.clear();
+  }
+
+  function handleMessage(message) {
+    if (message?.type === 'ready') {
+      if (message.ok === false) throw new Error(message.error || 'worker failed startup');
+      return;
+    }
+    const entry = pending.get(message.id);
+    if (!entry) return;
+    pending.delete(message.id);
+    clearTimeout(entry.timer);
+    if (message.ok) entry.resolve(message);
+    else entry.reject(new Error(message.error || 'worker request failed'));
+  }
+
+  function start() {
+    if (child && !child.killed) return readyPromise;
+    child = spawn(command, args, { stdio: ['pipe', 'pipe', 'pipe'] });
+    stdoutBuffer = '';
+    readyPromise = new Promise((resolve, reject) => {
+      const startupTimer = setTimeout(() => reject(new Error('worker startup timed out')), startupTimeoutMs);
+      const onData = chunk => {
+        stdoutBuffer += chunk.toString('utf8');
+        let index;
+        while ((index = stdoutBuffer.indexOf('\n')) >= 0) {
+          const line = stdoutBuffer.slice(0, index).trim();
+          stdoutBuffer = stdoutBuffer.slice(index + 1);
+          if (!line) continue;
+          let message;
+          try {
+            message = JSON.parse(line);
+          } catch (error) {
+            warn('mossttsnano_mlx worker emitted non-json stdout', line.slice(0, 300));
+            continue;
+          }
+          try {
+            handleMessage(message);
+            if (message?.type === 'ready') {
+              clearTimeout(startupTimer);
+              resolve();
+            }
+          } catch (error) {
+            clearTimeout(startupTimer);
+            reject(error);
+          }
+        }
+      };
+      child.stdout.on('data', onData);
+      child.stderr.on('data', chunk => warn('mossttsnano_mlx worker', chunk.toString('utf8').trim()));
+      child.on('error', error => {
+        clearTimeout(startupTimer);
+        reject(error);
+        rejectPending(error);
+      });
+      child.on('exit', (code, signal) => {
+        const error = new Error(`worker exited code=${code} signal=${signal}`);
+        child = null;
+        readyPromise = null;
+        clearTimeout(startupTimer);
+        rejectPending(error);
+      });
+    });
+    return readyPromise;
+  }
+
+  async function request(payload, { timeoutMs, signal } = {}) {
+    await start();
+    if (!child || !child.stdin.writable) throw new Error('worker stdin is not writable');
+    const id = nextId++;
+    const message = { id, ...payload };
+    return await new Promise((resolve, reject) => {
+      const timer = setTimeout(() => {
+        pending.delete(id);
+        reject(new Error('worker request timed out'));
+      }, timeoutMs || 180000);
+      const abort = () => {
+        clearTimeout(timer);
+        pending.delete(id);
+        reject(new Error('worker request aborted'));
+      };
+      if (signal) {
+        if (signal.aborted) return abort();
+        signal.addEventListener('abort', abort, { once: true });
+      }
+      pending.set(id, { resolve, reject, timer });
+      child.stdin.write(`${JSON.stringify(message)}\n`, error => {
+        if (error) {
+          clearTimeout(timer);
+          pending.delete(id);
+          reject(error);
+        }
+      });
+    });
+  }
+
+  function stop() {
+    if (child && !child.killed) child.kill('SIGTERM');
+    child = null;
+    readyPromise = null;
+  }
+
+  return { request, stop };
+}
+
+export function createMossTtsNanoMlxBackend(settings, deps = {}) {
+  const execFileAsync = deps.execFileAsync;
+  if (!execFileAsync) throw new Error('execFileAsync dependency is required');
+  const tmpdir = deps.tmpdir || os.tmpdir();
+  const warn = deps.warn || (() => {});
+  const fsApi = {
+    existsSync: deps.existsSync || fs.existsSync,
+    statSync: deps.statSync || fs.statSync,
+  };
+  const edge = createEdgeTtsBackend(settings, deps);
+  const mossttsnanoMlx = settings.mossttsnano_mlx;
+  const worker = mossttsnanoMlx.workerEnabled
+    ? createJsonLineWorker({
+        command: mossttsnanoMlx.python,
+        args: mossTtsNanoMlxWorkerArgs(mossttsnanoMlx),
+        spawn: deps.spawn || spawnProcess,
+        startupTimeoutMs: mossttsnanoMlx.workerStartupTimeoutMs,
+        warn,
+      })
+    : null;
+  return {
+    name: 'mossttsnano_mlx',
+    outputExtension: mossttsnanoMlx.useForProgress ? 'wav' : 'wav',
+    cacheKeyParts() {
+      return ['mossttsnano_mlx', mossttsnanoMlx.workerEnabled ? 'worker' : 'subprocess', mossttsnanoMlx.python, mossttsnanoMlx.script, mossttsnanoMlx.workerScript, mossttsnanoMlx.torchInferScript, mossttsnanoMlx.checkpoint, mossttsnanoMlx.audioTokenizer, mossttsnanoMlx.mode, mossttsnanoMlx.language, mossttsnanoMlx.torchDevice, mossttsnanoMlx.torchDtype, mossttsnanoMlx.promptAudio, mossttsnanoMlx.promptText, mossttsnanoMlx.maxNewFrames, mossttsnanoMlx.seed];
+    },
+    async synthesize(text, { signal, kind = 'final' } = {}) {
+      if (kind === 'progress' && !mossttsnanoMlx.useForProgress) {
+        return edge.synthesize(text, { signal, kind });
+      }
+      const out = uniquePath(tmpdir, 'verbalcoding-mossttsnano-mlx', 'wav');
+      try {
+        if (worker) {
+          await worker.request({ text, output_audio_path: out }, { timeoutMs: mossttsnanoMlx.timeoutMs, signal });
+        } else {
+          await execFileAsync(mossttsnanoMlx.python, mossTtsNanoMlxArgs(text, out, mossttsnanoMlx), execOptions({
+            timeout: mossttsnanoMlx.timeoutMs,
+            maxBuffer: 4 * 1024 * 1024,
+          }, signal));
+        }
+        return validateOutput(out, fsApi);
+      } catch (error) {
+        fs.rm(out, { force: true }, () => {});
+        notifyTtsFallback(deps, 'mossttsnano_mlx', error, kind);
+        return edge.synthesize(text, { signal, kind });
+      }
+    },
+    close() {
+      if (worker) worker.stop();
+    },
+  };
+}
+
 export function createTtsBackend(settings, deps = {}) {
   if (settings.backend === 'openvoice') return createOpenVoiceBackend(settings, deps);
   if (settings.backend === 'speechswift') return createSpeechSwiftBackend(settings, deps);
   if (settings.backend === 'supertonic') return createSupertonicBackend(settings, deps);
+  if (settings.backend === 'omnivoice') return createOmniVoiceBackend(settings, deps);
+  if (settings.backend === 'qwen3tts') return createQwen3TtsBackend(settings, deps);
+  if (settings.backend === 'mlxaudio') return createMlxAudioBackend(settings, deps);
+  if (settings.backend === 'neuttsair') return createNeuTtsAirBackend(settings, deps);
+  if (settings.backend === 'fireredtts2') return createFireRedTts2Backend(settings, deps);
+  if (settings.backend === 'mossttsnano') return createMossTtsNanoBackend(settings, deps);
+  if (settings.backend === 'mossttsnano_mlx') return createMossTtsNanoMlxBackend(settings, deps);
   return createEdgeTtsBackend(settings, deps);
 }